Ghassan Hamarneh

Saeid Asgari Taghanaki, Aliasghar Khani, Fereshte Khani et al. · stanford

A fundamental challenge of over-parameterized deep learning models is learning meaningful data representations that yield good performance on a downstream task without over-fitting spurious input features. This work proposes MaskTune, a masking strategy that prevents over-reliance on spurious (or a limited number of) features. MaskTune forces the trained model to explore new features during a single epoch finetuning by masking previously discovered features. MaskTune, unlike earlier approaches for mitigating shortcut learning, does not require any supervision, such as annotating spurious features or labels for subgroup samples in a dataset. Our empirical results on biased MNIST, CelebA, Waterbirds, and ImagenNet-9L datasets show that MaskTune is effective on tasks that often suffer from the existence of spurious correlations. Finally, we show that MaskTune outperforms or achieves similar performance to the competing methods when applied to the selective classification (classification with rejection option) task. Code for MaskTune is available at https://github.com/aliasgharkhani/Masktune.

Zahra Mirikharaji, Kumar Abhishek, Alceu Bissoto et al.

Skin cancer is a major public health problem that could benefit from computer-aided diagnosis to reduce the burden of this common disease. Skin lesion segmentation from images is an important step toward achieving this goal. However, the presence of natural and artificial artifacts (e.g., hair and air bubbles), intrinsic factors (e.g., lesion shape and contrast), and variations in image acquisition conditions make skin lesion segmentation a challenging task. Recently, various researchers have explored the applicability of deep learning models to skin lesion segmentation. In this survey, we cross-examine 177 research papers that deal with deep learning-based segmentation of skin lesions. We analyze these works along several dimensions, including input data (datasets, preprocessing, and synthetic data generation), model design (architecture, modules, and losses), and evaluation aspects (data annotation requirements and segmentation performance). We discuss these dimensions both from the viewpoint of select seminal works, and from a systematic viewpoint, examining how those choices have influenced current trends, and how their limitations should be addressed. To facilitate comparisons, we summarize all examined works in a comprehensive table as well as an interactive table available online at https://github.com/sfu-mial/skin-lesion-segmentation-survey.

Siyi Du, Nourhan Bayasi, Ghassan Hamarneh et al.

Despite its clinical utility, medical image segmentation (MIS) remains a daunting task due to images' inherent complexity and variability. Vision transformers (ViTs) have recently emerged as a promising solution to improve MIS; however, they require larger training datasets than convolutional neural networks. To overcome this obstacle, data-efficient ViTs were proposed, but they are typically trained using a single source of data, which overlooks the valuable knowledge that could be leveraged from other available datasets. Naivly combining datasets from different domains can result in negative knowledge transfer (NKT), i.e., a decrease in model performance on some domains with non-negligible inter-domain heterogeneity. In this paper, we propose MDViT, the first multi-domain ViT that includes domain adapters to mitigate data-hunger and combat NKT by adaptively exploiting knowledge in multiple small data resources (domains). Further, to enhance representation learning across domains, we integrate a mutual knowledge distillation paradigm that transfers knowledge between a universal network (spanning all the domains) and auxiliary domain-specific branches. Experiments on 4 skin lesion segmentation datasets show that MDViT outperforms state-of-the-art algorithms, with superior segmentation performance and a fixed model size, at inference time, even as more domains are added. Our code is available at https://github.com/siyi-wind/MDViT.

Siyi Du, Nourhan Bayasi, Ghassan Hamarneh et al.

Skin lesion segmentation (SLS) plays an important role in skin lesion analysis. Vision transformers (ViTs) are considered an auspicious solution for SLS, but they require more training data compared to convolutional neural networks (CNNs) due to their inherent parameter-heavy structure and lack of some inductive biases. To alleviate this issue, current approaches fine-tune pre-trained ViT backbones on SLS datasets, aiming to leverage the knowledge learned from a larger set of natural images to lower the amount of skin training data needed. However, fully fine-tuning all parameters of large backbones is computationally expensive and memory intensive. In this paper, we propose AViT, a novel efficient strategy to mitigate ViTs' data-hunger by transferring any pre-trained ViTs to the SLS task. Specifically, we integrate lightweight modules (adapters) within the transformer layers, which modulate the feature representation of a ViT without updating its pre-trained weights. In addition, we employ a shallow CNN as a prompt generator to create a prompt embedding from the input image, which grasps fine-grained information and CNN's inductive biases to guide the segmentation task on small datasets. Our quantitative experiments on 4 skin lesion datasets demonstrate that AViT achieves competitive, and at times superior, performance to SOTA but with significantly fewer trainable parameters. Our code is available at https://github.com/siyi-wind/AViT.

Nourhan Bayasi, Jamil Fayyad, Alceu Bissoto et al.

Continual Learning (CL) is crucial for enabling networks to dynamically adapt as they learn new tasks sequentially, accommodating new data and classes without catastrophic forgetting. Diverging from conventional perspectives on CL, our paper introduces a new perspective wherein forgetting could actually benefit the sequential learning paradigm. Specifically, we present BiasPruner, a CL framework that intentionally forgets spurious correlations in the training data that could lead to shortcut learning. Utilizing a new bias score that measures the contribution of each unit in the network to learning spurious features, BiasPruner prunes those units with the highest bias scores to form a debiased subnetwork preserved for a given task. As BiasPruner learns a new task, it constructs a new debiased subnetwork, potentially incorporating units from previous subnetworks, which improves adaptation and performance on the new task. During inference, BiasPruner employs a simple task-agnostic approach to select the best debiased subnetwork for predictions. We conduct experiments on three medical datasets for skin lesion classification and chest X-Ray classification and demonstrate that BiasPruner consistently outperforms SOTA CL methods in terms of classification performance and fairness. Our code is available here.

Siyi Du, Ben Hers, Nourhan Bayasi et al.

Deep learning models have achieved great success in automating skin lesion diagnosis. However, the ethnic disparity in these models' predictions, where lesions on darker skin types are usually underrepresented and have lower diagnosis accuracy, receives little attention. In this paper, we propose FairDisCo, a disentanglement deep learning framework with contrastive learning that utilizes an additional network branch to remove sensitive attributes, i.e. skin-type information from representations for fairness and another contrastive branch to enhance feature extraction. We compare FairDisCo to three fairness methods, namely, resampling, reweighting, and attribute-aware, on two newly released skin lesion datasets with different skin types: Fitzpatrick17k and Diverse Dermatology Images (DDI). We adapt two fairness-based metrics DPM and EOM for our multiple classes and sensitive attributes task, highlighting the skin-type bias in skin lesion classification. Extensive experimental evaluation demonstrates the effectiveness of FairDisCo, with fairer and superior performance on skin lesion classification tasks.

Kumar Abhishek, Ghassan Hamarneh

Training accurate medical image segmentation models requires large amounts of densely annotated data, which is costly and time-consuming to obtain. Semi-supervised learning (SSL) alleviates this by learning from both abundant unlabeled data and limited labeled data. However, most modern SSL methods rely on pseudolabels for unlabeled data, and typically assess their reliability through model confidence or uncertainty, measures that are self-referential and lack explicit grounding in segmentation quality. Instead, we propose a quality-guided SSL framework that trains a dedicated network to estimate segmentation quality from image-mask pairs. The predictor is trained on variable-quality masks generated through synthetic corruptions augmented with imperfect outputs from partially trained segmentation models, capturing realistic error patterns encountered during training. We integrate the quality predictor into SSL through two complementary mechanisms: a quality-aware regularization loss and a quality-based pseudolabel sample reweighting scheme. We show that our method serves as a drop-in enhancement to existing SSL frameworks. Extensive experiments across five datasets and multiple architectures demonstrate consistent improvements over competing SSL methods, advancing the state-of-the-art in semi-supervised medical image segmentation.

Arezou Pakzad, Kumar Abhishek, Ghassan Hamarneh

While deep learning based approaches have demonstrated expert-level performance in dermatological diagnosis tasks, they have also been shown to exhibit biases toward certain demographic attributes, particularly skin types (e.g., light versus dark), a fairness concern that must be addressed. We propose CIRCLe, a skin color invariant deep representation learning method for improving fairness in skin lesion classification. CIRCLe is trained to classify images by utilizing a regularization loss that encourages images with the same diagnosis but different skin types to have similar latent representations. Through extensive evaluation and ablation studies, we demonstrate CIRCLe's superior performance over the state-of-the-art when evaluated on 16k+ images spanning 6 Fitzpatrick skin types and 114 diseases, using classification accuracy, equal opportunity difference (for light versus dark groups), and normalized accuracy range, a new measure we propose to assess fairness on multiple skin type groups.

Mayur Mallya, Ghassan Hamarneh

Medical imaging is a cornerstone of therapy and diagnosis in modern medicine. However, the choice of imaging modality for a particular theranostic task typically involves trade-offs between the feasibility of using a particular modality (e.g., short wait times, low cost, fast acquisition, reduced radiation/invasiveness) and the expected performance on a clinical task (e.g., diagnostic accuracy, efficacy of treatment planning and guidance). In this work, we aim to apply the knowledge learned from the less feasible but better-performing (superior) modality to guide the utilization of the more-feasible yet under-performing (inferior) modality and steer it towards improved performance. We focus on the application of deep learning for image-based diagnosis. We develop a light-weight guidance model that leverages the latent representation learned from the superior modality, when training a model that consumes only the inferior modality. We examine the advantages of our method in the context of two clinical applications: multi-task skin lesion classification from clinical and dermoscopic images and brain tumor classification from multi-sequence magnetic resonance imaging (MRI) and histopathology images. For both these scenarios we show a boost in diagnostic performance of the inferior modality without requiring the superior modality. Furthermore, in the case of brain tumor classification, our method outperforms the model trained on the superior modality while producing comparable results to the model that uses both modalities during inference.

Kumar Abhishek, Jeremy Kawahara, Ghassan Hamarneh

Variability in medical image segmentation, arising from annotator preferences, expertise, and their choice of tools, has been well documented. While the majority of multi-annotator segmentation approaches focus on modeling annotator-specific preferences, they require annotator-segmentation correspondence. In this work, we introduce the problem of segmentation style discovery, and propose StyleSeg, a segmentation method that learns plausible, diverse, and semantically consistent segmentation styles from a corpus of image-mask pairs without any knowledge of annotator correspondence. StyleSeg consistently outperforms competing methods on four publicly available skin lesion segmentation (SLS) datasets. We also curate ISIC-MultiAnnot, the largest multi-annotator SLS dataset with annotator correspondence, and our results show a strong alignment, using our newly proposed measure AS2, between the predicted styles and annotator preferences. The code and the dataset are available at https://github.com/sfu-mial/StyleSeg.

Kumar Abhishek, Ghassan Hamarneh

While deep learning-based computer-aided diagnosis for skin lesion image analysis is approaching dermatologists' performance levels, there are several works showing that incorporating additional features such as shape priors, texture, color constancy, and illumination further improves the lesion diagnosis performance. In this work, we look at another clinically useful feature, skin lesion elevation, and investigate the feasibility of predicting and leveraging skin lesion elevation labels. Specifically, we use a deep learning model to predict image-level lesion elevation labels from 2D skin lesion images. We test the elevation prediction accuracy on the derm7pt dataset, and use the elevation prediction model to estimate elevation labels for images from five other datasets: ISIC 2016, 2017, and 2018 Challenge datasets, MSK, and DermoFit. We evaluate cross-domain generalization by using these estimated elevation labels as auxiliary inputs to diagnosis models, and show that these improve the classification performance, with AUROC improvements of up to 6.29% and 2.69% for dermoscopic and clinical images, respectively. The code is publicly available at https://github.com/sfu-mial/LesionElevation.

Weina Jin, Xiaoxiao Li, Ghassan Hamarneh

Being able to explain the prediction to clinical end-users is a necessity to leverage the power of artificial intelligence (AI) models for clinical decision support. For medical images, a feature attribution map, or heatmap, is the most common form of explanation that highlights important features for AI models' prediction. However, it is unknown how well heatmaps perform on explaining decisions on multi-modal medical images, where each image modality or channel visualizes distinct clinical information of the same underlying biomedical phenomenon. Understanding such modality-dependent features is essential for clinical users' interpretation of AI decisions. To tackle this clinically important but technically ignored problem, we propose the modality-specific feature importance (MSFI) metric. It encodes clinical image and explanation interpretation patterns of modality prioritization and modality-specific feature localization. We conduct a clinical requirement-grounded, systematic evaluation using computational methods and a clinician user study. Results show that the examined 16 heatmap algorithms failed to fulfill clinical requirements to correctly indicate AI model decision process or decision quality. The evaluation and MSFI metric can guide the design and selection of XAI algorithms to meet clinical requirements on multi-modal explanation.

Kumar Abhishek, Colin J. Brown, Ghassan Hamarneh

Modern deep learning training procedures rely on model regularization techniques such as data augmentation methods, which generate training samples that increase the diversity of data and richness of label information. A popular recent method, mixup, uses convex combinations of pairs of original samples to generate new samples. However, as we show in our experiments, mixup can produce undesirable synthetic samples, where the data is sampled off the manifold and can contain incorrect labels. We propose $ζ$-mixup, a generalization of mixup with provably and demonstrably desirable properties that allows convex combinations of $N \geq 2$ samples, leading to more realistic and diverse outputs that incorporate information from $N$ original samples by using a $p$-series interpolant. We show that, compared to mixup, $ζ$-mixup better preserves the intrinsic dimensionality of the original datasets, which is a desirable property for training generalizable models. Furthermore, we show that our implementation of $ζ$-mixup is faster than mixup, and extensive evaluation on controlled synthetic and 24 real-world natural and medical image classification datasets shows that $ζ$-mixup outperforms mixup and traditional data augmentation techniques.

Weina Jin, Jianyu Fan, Diane Gromala et al.

Non-technical end-users are silent and invisible users of the state-of-the-art explainable artificial intelligence (XAI) technologies. Their demands and requirements for AI explainability are not incorporated into the design and evaluation of XAI techniques, which are developed to explain the rationales of AI decisions to end-users and assist their critical decisions. This makes XAI techniques ineffective or even harmful in high-stakes applications, such as healthcare, criminal justice, finance, and autonomous driving systems. To systematically understand end-users' requirements to support the technical development of XAI, we conducted the EUCA user study with 32 layperson participants in four AI-assisted critical tasks. The study identified comprehensive user requirements for feature-, example-, and rule-based XAI techniques (manifested by the end-user-friendly explanation forms) and XAI evaluation objectives (manifested by the explanation goals), which were shown to be helpful to directly inspire the proposal of new XAI algorithms and evaluation metrics. The EUCA study findings, the identified explanation forms and goals for technical specification, and the EUCA study dataset support the design and evaluation of end-user-centered XAI techniques for accessible, safe, and accountable AI.

Weina Jin, Jianyu Fan, Diane Gromala et al.

The boundaries of existing explainable artificial intelligence (XAI) algorithms are confined to problems grounded in technical users' demand for explainability. This research paradigm disproportionately ignores the larger group of non-technical end users, who have a much higher demand for AI explanations in diverse explanation goals, such as making safer and better decisions and improving users' predicted outcomes. Lacking explainability-focused functional support for end users may hinder the safe and accountable use of AI in high-stakes domains, such as healthcare, criminal justice, finance, and autonomous driving systems. Built upon prior human factor analysis on end users' requirements for XAI, we identify and model four novel XAI technical problems covering the full spectrum from design to the evaluation of XAI algorithms, including edge-case-based reasoning, customizable counterfactual explanation, collapsible decision tree, and the verifiability metric to evaluate XAI utility. Based on these newly-identified research problems, we also discuss open problems in the technical development of user-centered XAI to inspire future research. Our work bridges human-centered XAI with the technical XAI community, and calls for a new research paradigm on the technical development of user-centered XAI for the responsible use of AI in critical tasks.

Weina Jin, Xiaoxiao Li, Ghassan Hamarneh

Explainable artificial intelligence (XAI) is motivated by the problem of making AI predictions understandable, transparent, and responsible, as AI becomes increasingly impactful in society and high-stakes domains. The evaluation and optimization criteria of XAI are gatekeepers for XAI algorithms to achieve their expected goals and should withstand rigorous inspection. To improve the scientific rigor of XAI, we conduct a critical examination of a common XAI criterion: plausibility. Plausibility assesses how convincing the AI explanation is to humans, and is usually quantified by metrics of feature localization or feature correlation. Our examination shows that plausibility is invalid to measure explainability, and human explanations are not the ground truth for XAI, because doing so ignores the necessary assumptions underpinning an explanation. Our examination further reveals the consequences of using plausibility as an XAI criterion, including increasing misleading explanations that manipulate users, deteriorating users' trust in the AI system, undermining human autonomy, being unable to achieve complementary human-AI task performance, and abandoning other possible approaches of enhancing understandability. Due to the invalidity of measurements and the unethical issues, this position paper argues that the community should stop using plausibility as a criterion for the evaluation and optimization of XAI algorithms. We also delineate new research approaches to improve XAI in trustworthiness, understandability, and utility to users, including complementary human-AI task performance.

Adriano D'Alessandro, Ali Mahdavi-Amiri, Ghassan Hamarneh

Crowd counting is a critical task in computer vision, with several important applications. However, existing counting methods rely on labor-intensive density map annotations, necessitating the manual localization of each individual pedestrian. While recent efforts have attempted to alleviate the annotation burden through weakly or semi-supervised learning, these approaches fall short of significantly reducing the workload. We propose a novel approach to eliminate the annotation burden by leveraging latent diffusion models to generate synthetic data. However, these models struggle to reliably understand object quantities, leading to noisy annotations when prompted to produce images with a specific quantity of objects. To address this, we use latent diffusion models to create two types of synthetic data: one by removing pedestrians from real images, which generates ranked image pairs with a weak but reliable object quantity signal, and the other by generating synthetic images with a predetermined number of objects, offering a strong but noisy counting signal. Our method utilizes the ranking image pairs for pre-training and then fits a linear layer to the noisy synthetic images using these crowd quantity features. We report state-of-the-art results for unsupervised crowd counting.

Kumar Abhishek, Jeremy Kawahara, Ghassan Hamarneh

Multi-annotator medical image segmentation is an important research problem, but requires annotated datasets that are expensive to collect. Dermoscopic skin lesion imaging allows human experts and AI systems to observe morphological structures otherwise not discernable from regular clinical photographs. However, currently there are no large-scale publicly available multi-annotator skin lesion segmentation (SLS) datasets with annotator-labels for dermoscopic skin lesion imaging. We introduce ISIC MultiAnnot++, a large public multi-annotator skin lesion segmentation dataset for images from the ISIC Archive. The final dataset contains 17,684 segmentation masks spanning 14,967 dermoscopic images, where 2,394 dermoscopic images have 2-5 segmentations per image, making it the largest publicly available SLS dataset. Further, metadata about the segmentation, including the annotators' skill level and segmentation tool, is included, enabling research on topics such as annotator-specific preference modeling for segmentation and annotator metadata analysis. We provide an analysis on the characteristics of this dataset, curated data partitions, and consensus segmentation masks.

Ashish Sinha, Ghassan Hamarneh

Anatomical trees play a central role in clinical diagnosis and treatment planning. However, accurately representing anatomical trees is challenging due to their varying and complex topology and geometry. Traditional methods for representing tree structures, captured using medical imaging, while invaluable for visualizing vascular and bronchial networks, exhibit drawbacks in terms of limited resolution, flexibility, and efficiency. Recently, implicit neural representations (INRs) have emerged as a powerful tool for representing shapes accurately and efficiently. We propose a novel approach, $TrIND$, for representing anatomical trees using INR, while also capturing the distribution of a set of trees via denoising diffusion in the space of INRs. We accurately capture the intricate geometries and topologies of anatomical trees at any desired resolution. Through extensive qualitative and quantitative evaluation, we demonstrate high-fidelity tree reconstruction with arbitrary resolution yet compact storage, and versatility across anatomical sites and tree complexities. The code is available at: \texttt{\url{https://github.com/sinashish/TreeDiffusion}}.

Aliasghar Khani, Saeid Asgari Taghanaki, Aditya Sanghi et al.

Significant strides have been made using large vision-language models, like Stable Diffusion (SD), for a variety of downstream tasks, including image editing, image correspondence, and 3D shape generation. Inspired by these advancements, we explore leveraging these extensive vision-language models for segmenting images at any desired granularity using as few as one annotated sample by proposing SLiMe. SLiMe frames this problem as an optimization task. Specifically, given a single training image and its segmentation mask, we first extract attention maps, including our novel "weighted accumulated self-attention map" from the SD prior. Then, using the extracted attention maps, the text embeddings of Stable Diffusion are optimized such that, each of them, learn about a single segmented region from the training image. These learned embeddings then highlight the segmented region in the attention maps, which in turn can then be used to derive the segmentation map. This enables SLiMe to segment any real-world image during inference with the granularity of the segmented region in the training image, using just one example. Moreover, leveraging additional training data when available, i.e. few-shot, improves the performance of SLiMe. We carried out a knowledge-rich set of experiments examining various design factors and showed that SLiMe outperforms other existing one-shot and few-shot segmentation methods.

Nourhan Bayasi, Jamil Fayyad, Ghassan Hamarneh et al.

Simplicity bias poses a significant challenge in neural networks, often leading models to favor simpler solutions and inadvertently learn decision rules influenced by spurious correlations. This results in biased models with diminished generalizability. While many current approaches depend on human supervision, obtaining annotations for various bias attributes is often impractical. To address this, we introduce Debiasify, a novel self-distillation approach that requires no prior knowledge about the nature of biases. Our method leverages a new distillation loss to transfer knowledge within the network, from deeper layers containing complex, highly-predictive features to shallower layers with simpler, attribute-conditioned features in an unsupervised manner. This enables Debiasify to learn robust, debiased representations that generalize effectively across diverse biases and datasets, improving both worst-group performance and overall accuracy. Extensive experiments on computer vision and medical imaging benchmarks demonstrate the effectiveness of our approach, significantly outperforming previous unsupervised debiasing methods (e.g., a 10.13% improvement in worst-group accuracy for Wavy Hair classification in CelebA) and achieving comparable or superior performance to supervised approaches. Our code is publicly available at the following link: Debiasify.

Kumar Abhishek, Jeremy Kawahara, Ghassan Hamarneh

Medical image segmentation exhibits intra- and inter-annotator variability due to ambiguous object boundaries, annotator preferences, expertise, and tools, among other factors. Lesions with ambiguous boundaries, e.g., spiculated or infiltrative nodules, or irregular borders per the ABCD rule, are particularly prone to disagreement and are often associated with malignancy. In this work, we curate IMA++, the largest multi-annotator skin lesion segmentation dataset, on which we conduct an in-depth study of variability due to annotator, malignancy, tool, and skill factors. We find a statistically significant (p<0.001) association between inter-annotator agreement (IAA), measured using Dice, and the malignancy of skin lesions. We further show that IAA can be accurately predicted directly from dermoscopic images, achieving a mean absolute error of 0.108. Finally, we leverage this association by utilizing IAA as a "soft" clinical feature within a multi-task learning objective, yielding a 4.2% improvement in balanced accuracy averaged across multiple model architectures and across IMA++ and four public dermoscopic datasets. The code is available at https://github.com/sfu-mial/skin-IAV.

Ashish Sinha, Jeremy Kawahara, Arezou Pakzad et al.

In recent years, deep learning (DL) has shown great potential in the field of dermatological image analysis. However, existing datasets in this domain have significant limitations, including a small number of image samples, limited disease conditions, insufficient annotations, and non-standardized image acquisitions. To address these shortcomings, we propose a novel framework called DermSynth3D. DermSynth3D blends skin disease patterns onto 3D textured meshes of human subjects using a differentiable renderer and generates 2D images from various camera viewpoints under chosen lighting conditions in diverse background scenes. Our method adheres to top-down rules that constrain the blending and rendering process to create 2D images with skin conditions that mimic in-the-wild acquisitions, ensuring more meaningful results. The framework generates photo-realistic 2D dermoscopy images and the corresponding dense annotations for semantic segmentation of the skin, skin conditions, body parts, bounding boxes around lesions, depth maps, and other 3D scene parameters, such as camera position and lighting conditions. DermSynth3D allows for the creation of custom datasets for various dermatology tasks. We demonstrate the effectiveness of data generated using DermSynth3D by training DL models on synthetic data and evaluating them on various dermatology tasks using real 2D dermatological images. We make our code publicly available at https://github.com/sfu-mial/DermSynth3D.

M. Sadegh Saberian, Kathleen P. Moriarty, Andrea D. Olmstead et al.

Drug repurposing can accelerate the identification of effective compounds for clinical use against SARS-CoV-2, with the advantage of pre-existing clinical safety data and an established supply chain. RNA viruses such as SARS-CoV-2 manipulate cellular pathways and induce reorganization of subcellular structures to support their life cycle. These morphological changes can be quantified using bioimaging techniques. In this work, we developed DEEMD: a computational pipeline using deep neural network models within a multiple instance learning framework, to identify putative treatments effective against SARS-CoV-2 based on morphological analysis of the publicly available RxRx19a dataset. This dataset consists of fluorescence microscopy images of SARS-CoV-2 non-infected cells and infected cells, with and without drug treatment. DEEMD first extracts discriminative morphological features to generate cell morphological profiles from the non-infected and infected cells. These morphological profiles are then used in a statistical model to estimate the applied treatment efficacy on infected cells based on similarities to non-infected cells. DEEMD is capable of localizing infected cells via weak supervision without any expensive pixel-level annotations. DEEMD identifies known SARS-CoV-2 inhibitors, such as Remdesivir and Aloxistatin, supporting the validity of our approach. DEEMD can be explored for use on other emerging viruses and datasets to rapidly identify candidate antiviral treatments in the future}. Our implementation is available online at https://www.github.com/Sadegh-Saberian/DEEMD

Kumar Abhishek, Ghassan Hamarneh

The segmentation of skin lesions is a crucial task in clinical decision support systems for the computer aided diagnosis of skin lesions. Although deep learning-based approaches have improved segmentation performance, these models are often susceptible to class imbalance in the data, particularly, the fraction of the image occupied by the background healthy skin. Despite variations of the popular Dice loss function being proposed to tackle the class imbalance problem, the Dice loss formulation does not penalize misclassifications of the background pixels. We propose a novel metric-based loss function using the Matthews correlation coefficient, a metric that has been shown to be efficient in scenarios with skewed class distributions, and use it to optimize deep segmentation models. Evaluations on three skin lesion image datasets: the ISBI ISIC 2017 Skin Lesion Segmentation Challenge dataset, the DermoFit Image Library, and the PH2 dataset, show that models trained using the proposed loss function outperform those trained using Dice loss by 11.25%, 4.87%, and 0.76% respectively in the mean Jaccard index. The code is available at https://github.com/kakumarabhishek/MCC-Loss.

Amirhossein Alimohammadi, Sauradip Nag, Saeid Asgari Taghanaki et al.

Segmenting an object in a video presents significant challenges. Each pixel must be accurately labelled, and these labels must remain consistent across frames. The difficulty increases when the segmentation is with arbitrary granularity, meaning the number of segments can vary arbitrarily, and masks are defined based on only one or a few sample images. In this paper, we address this issue by employing a pre-trained text to image diffusion model supplemented with an additional tracking mechanism. We demonstrate that our approach can effectively manage various segmentation scenarios and outperforms state-of-the-art alternatives.

Adriano D'Alessandro, Ali Mahdavi-Amiri, Ghassan Hamarneh

Object counting methods typically rely on manually annotated datasets. The cost of creating such datasets has restricted the versatility of these networks to count objects from specific classes (such as humans or penguins), and counting objects from diverse categories remains a challenge. The availability of robust text-to-image latent diffusion models (LDMs) raises the question of whether these models can be utilized to generate counting datasets. However, LDMs struggle to create images with an exact number of objects based solely on text prompts but they can be used to offer a dependable \textit{sorting} signal by adding and removing objects within an image. Leveraging this data, we initially introduce an unsupervised sorting methodology to learn object-related features that are subsequently refined and anchored for counting purposes using counting data generated by LDMs. Further, we present a density classifier-guided method for dividing an image into patches containing objects that can be reliably counted. Consequently, we can generate counting data for any type of object and count them in an unsupervised manner. Our approach outperforms other unsupervised and few-shot alternatives and is not restricted to specific object classes for which counting data is available. Code to be released upon acceptance.

Weina Jin, Nicholas Vincent, Ghassan Hamarneh

"why" we develop AI. Lacking critical reflections on the general visions and purposes of AI may make the community vulnerable to manipulation. In this position paper, we explore the "why" question of AI. We denote answers to the "why" question the imaginations of AI, which depict our general visions, frames, and mindsets for the prospects of AI. We identify that the prevailing vision in the AI community is largely a monoculture that emphasizes objectives such as replacing humans and improving productivity. Our critical examination of this mainstream imagination highlights its underpinning and potentially unjust assumptions. We then call to diversify our collective imaginations of AI, embedding ethical assumptions from the outset in the imaginations of AI. To facilitate the community's pursuit of diverse imaginations, we demonstrate one process for constructing a new imagination of "AI for just work," and showcase its application in the medical image synthesis task to make it more ethical. We hope this work will help the AI community to open critical dialogues with civil society on the visions and purposes of AI, and inspire more technical works and advocacy in pursuit of diverse and ethical imaginations to restore the value of AI for the public good.

Weina Jin, Ashish Sinha, Kumar Abhishek et al.

The task of ethical Medical Image Synthesis (MISyn) is to ensure that the MISyn techniques are researched and developed ethically throughout their entire lifecycle, which is essential to prevent the negative impacts of MISyn. To address the ever-increasing needs and requirements for ethical practice of MISyn research and development, we first conduct a theoretical analysis that identifies the key properties of ethical MISyn and intrinsic limits of MISyn. We identify that synthetic images lack inherent grounding in real medical phenomena, cannot fully represent the training medical images, and inevitably introduce new distribution shifts and biases. Ethical risks can arise from not acknowledging the intrinsic limits and weaknesses of synthetic images compared to medical images, with the extreme form manifested as misinformation of MISyn that substitutes synthetic images for medical images without acknowledgment. The resulting ethical harms include eroding trust in the medical imaging dataset environment and causing algorithmic discrimination towards stakeholders and the public. To facilitate collective efforts towards ethical MISyn within and outside the medical image analysis community, we then propose practical supports for ethical practice in MISyn based on the theoretical analysis, including ethical practice recommendations that adapt the existing technical standards, problem formulation, design, and evaluation practice of MISyn to the ethical challenges; and oversight recommendations to facilitate checks and balances from stakeholders and the public. We also present two case studies that demonstrate how to apply the ethical practice recommendations in practice, and identify gaps between existing practice and the ethical practice recommendations.

Adriano D'Alessandro, Ali Mahdavi-Amiri, Ghassan Hamarneh

Fine-grained object counting remains a major challenge for class-agnostic counting models, which overcount visually similar but incorrect instances (e.g., jalapeño vs. poblano). Addressing this by annotating new data and fully retraining the model is time-consuming and does not guarantee generalization to additional novel categories at test time. Instead, we propose an alternative paradigm: Given a category name, tune a compact concept embedding derived from the prompt using synthetic images and pseudo-labels generated by a text-to-image diffusion model. This embedding conditions a specialization module that refines raw overcounts from any frozen counter into accurate, category-specific estimates\textemdash without requiring real images or human annotations. We validate our approach on \textsc{Lookalikes}, a challenging new benchmark containing 1,037 images across 27 fine-grained subcategories, and show substantial improvements over strong baselines. Code will be released upon acceptance. Dataset - https://dalessandro.dev/datasets/lookalikes/

Tanya Gatsak, Kumar Abhishek, Hanene Ben Yedder et al.

PET imaging is an invaluable tool in clinical settings as it captures the functional activity of both healthy anatomy and cancerous lesions. Developing automatic lesion segmentation methods for PET images is crucial since manual lesion segmentation is laborious and prone to inter- and intra-observer variability. We propose PET-Disentangler, a 3D disentanglement method that uses a 3D UNet-like encoder-decoder architecture to disentangle disease and normal healthy anatomical features with losses for segmentation, reconstruction, and healthy component plausibility. A critic network is used to encourage the healthy latent features to match the distribution of healthy samples and thus encourages these features to not contain any lesion-related features. Our quantitative results show that PET-Disentangler is less prone to incorrectly declaring healthy and high tracer uptake regions as cancerous lesions, since such uptake pattern would be assigned to the disentangled healthy component.

Kumar Abhishek, Aditi Jain, Ghassan Hamarneh

The remarkable progress of deep learning in dermatological tasks has brought us closer to achieving diagnostic accuracies comparable to those of human experts. However, while large datasets play a crucial role in the development of reliable deep neural network models, the quality of data therein and their correct usage are of paramount importance. Several factors can impact data quality, such as the presence of duplicates, data leakage across train-test partitions, mislabeled images, and the absence of a well-defined test partition. In this paper, we conduct meticulous analyses of three popular dermatological image datasets: DermaMNIST, its source HAM10000, and Fitzpatrick17k, uncovering these data quality issues, measure the effects of these problems on the benchmark results, and propose corrections to the datasets. Besides ensuring the reproducibility of our analysis, by making our analysis pipeline and the accompanying code publicly available, we aim to encourage similar explorations and to facilitate the identification and addressing of potential data quality issues in other large datasets.

Ivan R. Nabi, Ben Cardoen, Ismail M. Khater et al.

Super-resolution microscopy, or nanoscopy, enables the use of fluorescent-based molecular localization tools to study molecular structure at the nanoscale level in the intact cell, bridging the mesoscale gap to classical structural biology methodologies. Analysis of super-resolution data by artificial intelligence (AI), such as machine learning, offers tremendous potential for discovery of new biology, that, by definition, is not known and lacks ground truth. Herein, we describe the application of weakly supervised paradigms to super-resolution microscopy and its potential to enable the accelerated exploration of the nanoscale architecture of subcellular macromolecules and organelles.

Weina Jin, Xiaoxiao Li, Mostafa Fatehi et al.

Explainable artificial intelligence (XAI) is essential for enabling clinical users to get informed decision support from AI and comply with evidence-based medical practice. Applying XAI in clinical settings requires proper evaluation criteria to ensure the explanation technique is both technically sound and clinically useful, but specific support is lacking to achieve this goal. To bridge the research gap, we propose the Clinical XAI Guidelines that consist of five criteria a clinical XAI needs to be optimized for. The guidelines recommend choosing an explanation form based on Guideline 1 (G1) Understandability and G2 Clinical relevance. For the chosen explanation form, its specific XAI technique should be optimized for G3 Truthfulness, G4 Informative plausibility, and G5 Computational efficiency. Following the guidelines, we conducted a systematic evaluation on a novel problem of multi-modal medical image explanation with two clinical tasks, and proposed new evaluation metrics accordingly. Sixteen commonly-used heatmap XAI techniques were evaluated and found to be insufficient for clinical use due to their failure in G3 and G4. Our evaluation demonstrated the use of Clinical XAI Guidelines to support the design and evaluation of clinically viable XAI.

Weina Jin, Xiaoxiao Li, Ghassan Hamarneh

Being able to explain the prediction to clinical end-users is a necessity to leverage the power of AI models for clinical decision support. For medical images, saliency maps are the most common form of explanation. The maps highlight important features for AI model's prediction. Although many saliency map methods have been proposed, it is unknown how well they perform on explaining decisions on multi-modal medical images, where each modality/channel carries distinct clinical meanings of the same underlying biomedical phenomenon. Understanding such modality-dependent features is essential for clinical users' interpretation of AI decisions. To tackle this clinically important but technically ignored problem, we propose the MSFI (Modality-Specific Feature Importance) metric to examine whether saliency maps can highlight modality-specific important features. MSFI encodes the clinical requirements on modality prioritization and modality-specific feature localization. Our evaluations on 16 commonly used saliency map methods, including a clinician user study, show that although most saliency map methods captured modality importance information in general, most of them failed to highlight modality-specific important features consistently and precisely. The evaluation results guide the choices of saliency map methods and provide insights to propose new ones targeting clinical applications.

Mengliu Zhao, Jeremy Kawahara, Kumar Abhishek et al.

We present an automated approach to detect and longitudinally track skin lesions on 3D total-body skin surface scans. The acquired 3D mesh of the subject is unwrapped to a 2D texture image, where a trained objected detection model, Faster R-CNN, localizes the lesions within the 2D domain. These detected skin lesions are mapped back to the 3D surface of the subject and, for subjects imaged multiple times, we construct a graph-based matching procedure to longitudinally track lesions that considers the anatomical correspondences among pairs of meshes and the geodesic proximity of corresponding lesions and the inter-lesion geodesic distances. We evaluated the proposed approach using 3DBodyTex, a publicly available dataset composed of 3D scans imaging the coloured skin (textured meshes) of 200 human subjects. We manually annotated locations that appeared to the human eye to contain a pigmented skin lesion as well as tracked a subset of lesions occurring on the same subject imaged in different poses. Our results, when compared to three human annotators, suggest that the trained Faster R-CNN detects lesions at a similar performance level as the human annotators. Our lesion tracking algorithm achieves an average matching accuracy of 88% on a set of detected corresponding pairs of prominent lesions of subjects imaged in different poses, and an average longitudinal accuracy of 71% when encompassing additional errors due to lesion detection. As there currently is no other large-scale publicly available dataset of 3D total-body skin lesions, we publicly release over 25,000 3DBodyTex manual annotations, which we hope will further research on total-body skin lesion analysis.

Weina Jin, Jianyu Fan, Diane Gromala et al.

The ability to explain decisions to end-users is a necessity to deploy AI as critical decision support. Yet making AI explainable to non-technical end-users is a relatively ignored and challenging problem. To bridge the gap, we first identify twelve end-user-friendly explanatory forms that do not require technical knowledge to comprehend, including feature-, example-, and rule-based explanations. We then instantiate the explanatory forms as prototyping cards in four AI-assisted critical decision-making tasks, and conduct a user study to co-design low-fidelity prototypes with 32 layperson participants. The results confirm the relevance of using explanatory forms as building blocks of explanations, and identify their proprieties - pros, cons, applicable explanation goals, and design implications. The explanatory forms, their proprieties, and prototyping supports (including a suggested prototyping process, design templates and exemplars, and associated algorithms to actualize explanatory forms) constitute the End-User-Centered explainable AI framework EUCA, and is available at http://weinajin.github.io/end-user-xai . It serves as a practical prototyping toolkit for HCI/AI practitioners and researchers to understand user requirements and build end-user-centered explainable AI.

Zahra Mirikharaji, Kumar Abhishek, Saeed Izadi et al.

Medical image segmentation annotations suffer from inter- and intra-observer variations even among experts due to intrinsic differences in human annotators and ambiguous boundaries. Leveraging a collection of annotators' opinions for an image is an interesting way of estimating a gold standard. Although training deep models in a supervised setting with a single annotation per image has been extensively studied, generalizing their training to work with datasets containing multiple annotations per image remains a fairly unexplored problem. In this paper, we propose an approach to handle annotators' disagreements when training a deep model. To this end, we propose an ensemble of Bayesian fully convolutional networks (FCNs) for the segmentation task by considering two major factors in the aggregation of multiple ground truth annotations: (1) handling contradictory annotations in the training data originating from inter-annotator disagreements and (2) improving confidence calibration through the fusion of base models' predictions. We demonstrate the superior performance of our approach on the ISIC Archive and explore the generalization performance of our proposed method by cross-dataset evaluation on the PH2 and DermoFit datasets.

Saeed Izadi, Ghassan Hamarneh

Confocal microscopy is essential for histopathologic cell visualization and quantification. Despite its significant role in biology, fluorescence confocal microscopy suffers from the presence of inherent noise during image acquisition. Non-local patch-wise Bayesian mean filtering (NLB) was until recently the state-of-the-art denoising approach. However, classic denoising methods have been outperformed by neural networks in recent years. In this work, we propose to exploit the strengths of NLB in the framework of Bayesian deep learning. We do so by designing a convolutional neural network and training it to learn parameters of a Gaussian model approximating the prior on noise-free patches given their nearest, similar yet non-local, neighbors. We then apply Bayesian reasoning to leverage the prior and information from the noisy patch in the process of approximating the noise-free patch. Specifically, we use the closed-form analytic \textit{maximum a posteriori} (MAP) estimate in the NLB algorithm to obtain the noise-free patch that maximizes the posterior distribution. The performance of our proposed method is evaluated on confocal microscopy images with real noise Poisson-Gaussian noise. Our experiments reveal the superiority of our approach against state-of-the-art unsupervised denoising techniques.

Kumar Abhishek, Ghassan Hamarneh, Mark S. Drew

The semantic segmentation of skin lesions is an important and common initial task in the computer aided diagnosis of dermoscopic images. Although deep learning-based approaches have considerably improved the segmentation accuracy, there is still room for improvement by addressing the major challenges, such as variations in lesion shape, size, color and varying levels of contrast. In this work, we propose the first deep semantic segmentation framework for dermoscopic images which incorporates, along with the original RGB images, information extracted using the physics of skin illumination and imaging. In particular, we incorporate information from specific color bands, illumination invariant grayscale images, and shading-attenuated images. We evaluate our method on three datasets: the ISBI ISIC 2017 Skin Lesion Segmentation Challenge dataset, the DermoFit Image Library, and the PH2 dataset and observe improvements of 12.02%, 4.30%, and 8.86% respectively in the mean Jaccard index over a baseline model trained only with RGB images.

Hanene Ben Yedder, Ben Cardoen, Ghassan Hamarneh

Medical imaging is an invaluable resource in medicine as it enables to peer inside the human body and provides scientists and physicians with a wealth of information indispensable for understanding, modelling, diagnosis, and treatment of diseases. Reconstruction algorithms entail transforming signals collected by acquisition hardware into interpretable images. Reconstruction is a challenging task given the ill-posed of the problem and the absence of exact analytic inverse transforms in practical cases. While the last decades witnessed impressive advancements in terms of new modalities, improved temporal and spatial resolution, reduced cost, and wider applicability, several improvements can still be envisioned such as reducing acquisition and reconstruction time to reduce patient's exposure to radiation and discomfort while increasing clinics throughput and reconstruction accuracy. Furthermore, the deployment of biomedical imaging in handheld devices with small power requires a fine balance between accuracy and latency.

Weina Jin, Mostafa Fatehi, Kumar Abhishek et al.

Primary brain tumors including gliomas continue to pose significant management challenges to clinicians. While the presentation, the pathology, and the clinical course of these lesions are variable, the initial investigations are usually similar. Patients who are suspected to have a brain tumor will be assessed with computed tomography (CT) and magnetic resonance imaging (MRI). The imaging findings are used by neurosurgeons to determine the feasibility of surgical resection and plan such an undertaking. Imaging studies are also an indispensable tool in tracking tumor progression or its response to treatment. As these imaging studies are non-invasive, relatively cheap and accessible to patients, there have been many efforts over the past two decades to increase the amount of clinically-relevant information that can be extracted from brain imaging. Most recently, artificial intelligence (AI) techniques have been employed to segment and characterize brain tumors, as well as to detect progression or treatment-response. However, the clinical utility of such endeavours remains limited due to challenges in data collection and annotation, model training, and the reliability of AI-generated information. We provide a review of recent advances in addressing the above challenges. First, to overcome the challenge of data paucity, different image imputation and synthesis techniques along with annotation collection efforts are summarized. Next, various training strategies are presented to meet multiple desiderata, such as model performance, generalization ability, data privacy protection, and learning with sparse annotations. Finally, standardized performance evaluation and model interpretability methods have been reviewed. We believe that these technical approaches will facilitate the development of a fully-functional AI tool in the clinical care of patients with gliomas.

Saeid Asgari Taghanaki, Kumar Abhishek, Ghassan Hamarneh

Over-parameterized deep models usually over-fit to a given training distribution, which makes them sensitive to small changes and out-of-distribution samples at inference time, leading to low generalization performance. To this end, several model-based and randomized data-dependent regularization methods are applied, such as data augmentation, which prevents a model from memorizing the training distribution. Instead of the random transformation of the input images, we propose SIGN, a new regularization method, which modifies the input variables using a linear transformation by estimating each variable's contribution to the final prediction. Our proposed technique maps the input data to a new manifold where the less important variables are de-emphasized. To test the effectiveness of the proposed idea and compare it with other competing methods, we design several test scenarios, such as classification performance, uncertainty, out-of-distribution, and robustness analyses. We compare the methods using three different datasets and four models. We find that SIGN encourages more compact class representations, which results in the model's robustness to random corruptions and out-of-distribution samples while also simultaneously achieving superior performance on normal data compared to other competing methods. Our experiments also demonstrate the successful transferability of the SIGN samples from one model to another.

Aïcha BenTaieb, Ghassan Hamarneh

Histopathology images; microscopy images of stained tissue biopsies contain fundamental prognostic information that forms the foundation of pathological analysis and diagnostic medicine. However, diagnostics from histopathology images generally rely on a visual cognitive assessment of tissue slides which implies an inherent element of interpretation and hence subjectivity. Access to digitized histopathology images enabled the development of computational systems aiming at reducing manual intervention and automating parts of pathologists' workflow. Specifically, applications of deep learning to histopathology image analysis now offer opportunities for better quantitative modeling of disease appearance and hence possibly improved prediction of disease aggressiveness and patient outcome. However digitized histopathology tissue slides are unique in a variety of ways and come with their own set of computational challenges. In this survey, we summarize the different challenges facing computational systems for digital pathology and provide a review of state-of-the-art works that developed deep learning-based solutions for the predictive modeling of histopathology images from a detection, stain normalization, segmentation, and tissue classification perspective. We then discuss the challenges facing the validation and integration of such deep learning-based computational systems in clinical workflow and reflect on future opportunities for histopathology derived image measurements and better predictive modeling.

Shahab Aslani, Vittorio Murino, Michael Dayan et al.

This paper presents a simple and effective generalization method for magnetic resonance imaging (MRI) segmentation when data is collected from multiple MRI scanning sites and as a consequence is affected by (site-)domain shifts. We propose to integrate a traditional encoder-decoder network with a regularization network. This added network includes an auxiliary loss term which is responsible for the reduction of the domain shift problem and for the resulting improved generalization. The proposed method was evaluated on multiple sclerosis lesion segmentation from MRI data. We tested the proposed model on an in-house clinical dataset including 117 patients from 56 different scanning sites. In the experiments, our method showed better generalization performance than other baseline networks.

Saeid Asgari Taghanaki, Kumar Abhishek, Joseph Paul Cohen et al.

The semantic image segmentation task consists of classifying each pixel of an image into an instance, where each instance corresponds to a class. This task is a part of the concept of scene understanding or better explaining the global context of an image. In the medical image analysis domain, image segmentation can be used for image-guided interventions, radiotherapy, or improved radiological diagnostics. In this review, we categorize the leading deep learning-based medical and non-medical image segmentation solutions into six main groups of deep architectural, data synthesis-based, loss function-based, sequenced models, weakly supervised, and multi-task methods and provide a comprehensive review of the contributions in each of these groups. Further, for each group, we analyze each variant of these groups and discuss the limitations of the current approaches and present potential future research directions for semantic image segmentation.

Saeed Izadi, Zahra Mirikharaji, Mengliu Zhao et al.

The accuracy of medical imaging-based diagnostics is directly impacted by the quality of the collected images. A passive approach to improve image quality is one that lags behind improvements in imaging hardware, awaiting better sensor technology of acquisition devices. An alternative, active strategy is to utilize prior knowledge of the imaging system to directly post-process and improve the acquired images. Traditionally, priors about the image properties are taken into account to restrict the solution space. However, few techniques exploit the prior about the noise properties. In this paper, we propose a neural network-based model for disentangling the signal and noise components of an input noisy image, without the need for any ground truth training data. We design a unified loss function that encodes priors about signal as well as noise estimate in the form of regularization terms. Specifically, by using total variation and piecewise constancy priors along with noise whiteness priors such as auto-correlation and stationary losses, our network learns to decouple an input noisy image into the underlying signal and noise components. We compare our proposed method to Noise2Noise and Noise2Self, as well as non-local mean and BM3D, on three public confocal laser endomicroscopy datasets. Experimental results demonstrate the superiority of our network compared to state-of-the-art in terms of PSNR and SSIM.

Saeed Izadi, Darren Sutton, Ghassan Hamarneh

Recent developments in image acquisition literature have miniaturized the confocal laser endomicroscopes to improve usability and flexibility of the apparatus in actual clinical settings. However, miniaturized devices collect less light and have fewer optical components, resulting in pixelation artifacts and low resolution images. Owing to the strength of deep networks, many supervised methods known as super resolution have achieved considerable success in restoring low resolution images by generating the missing high frequency details. In this work, we propose a novel attention mechanism that, for the first time, combines 1st- and 2nd-order statistics for pooling operation, in the spatial and channel-wise dimensions. We compare the efficacy of our method to 11 other existing single image super resolution techniques that compensate for the reduction in image quality caused by the necessity of endomicroscope miniaturization. All evaluations are carried out on three publicly available datasets. Experimental results show that our method can produce competitive results against state-of-the-art in terms of PSNR, SSIM, and IFC metrics. Additionally, our proposed method contains small number of parameters, which makes it lightweight and fast for real-time applications.

Kumar Abhishek, Ghassan Hamarneh

Skin lesion segmentation is a vital task in skin cancer diagnosis and further treatment. Although deep learning based approaches have significantly improved the segmentation accuracy, these algorithms are still reliant on having a large enough dataset in order to achieve adequate results. Inspired by the immense success of generative adversarial networks (GANs), we propose a GAN-based augmentation of the original dataset in order to improve the segmentation performance. In particular, we use the segmentation masks available in the training dataset to train the Mask2Lesion model, and use the model to generate new lesion images given any arbitrary mask, which are then used to augment the original training dataset. We test Mask2Lesion augmentation on the ISBI ISIC 2017 Skin Lesion Segmentation Challenge dataset and achieve an improvement of 5.17% in the mean Dice score as compared to a model trained with only classical data augmentation techniques.

Zahra Mirikharaji, Yiqi Yan, Ghassan Hamarneh

Deep convolutional neural networks have driven substantial advancements in the automatic understanding of images. Requiring a large collection of images and their associated annotations is one of the main bottlenecks limiting the adoption of deep networks. In the task of medical image segmentation, requiring pixel-level semantic annotations performed by human experts exacerbate this difficulty. This paper proposes a new framework to train a fully convolutional segmentation network from a large set of cheap unreliable annotations and a small set of expert-level clean annotations. We propose a spatially adaptive reweighting approach to treat clean and noisy pixel-level annotations commensurately in the loss function. We deploy a meta-learning approach to assign higher importance to pixels whose loss gradient direction is closer to those of clean data. Our experiments on training the network using segmentation ground truth corrupted with different levels of annotation noise show how spatial reweighting improves the robustness of deep networks to noisy annotations.