Lin Yang

Honglin Li, Yunlong Zhang, Pingyi Chen et al.

Histopathology Whole Slide Image (WSI) analysis serves as the gold standard for clinical cancer diagnosis in the daily routines of doctors. To develop computer-aided diagnosis model for WSIs, previous methods typically employ Multi-Instance Learning to enable slide-level prediction given only slide-level labels. Among these models, vanilla attention mechanisms without pairwise interactions have traditionally been employed but are unable to model contextual information. More recently, self-attention models have been utilized to address this issue. To alleviate the computational complexity of long sequences in large WSIs, methods like HIPT use region-slicing, and TransMIL employs approximation of full self-attention. Both approaches suffer from suboptimal performance due to the loss of key information. Moreover, their use of absolute positional embedding struggles to effectively handle long contextual dependencies in shape-varying WSIs. In this paper, we first analyze how the low-rank nature of the long-sequence attention matrix constrains the representation ability of WSI modelling. Then, we demonstrate that the rank of attention matrix can be improved by focusing on local interactions via a local attention mask. Our analysis shows that the local mask aligns with the attention patterns in the lower layers of the Transformer. Furthermore, the local attention mask can be implemented during chunked attention calculation, reducing the quadratic computational complexity to linear with a small local bandwidth. Building on this, we propose a local-global hybrid Transformer for both computational acceleration and local-global information interactions modelling. Our method, Long-contextual MIL (LongMIL), is evaluated through extensive experiments on various WSI tasks to validate its superiority. Our code will be available at github.com/invoker-LL/Long-MIL.

Minghao Tang, Yongquan He, Yongxiu Xu et al.

Named entity recognition (NER) is a fundamental task in natural language processing that aims to identify and classify named entities in text. However, span-based methods for NER typically assign entity types to text spans, resulting in an imbalanced sample space and neglecting the connections between non-entity and entity spans. To address these issues, we propose a novel approach for NER, named the Boundary Offset Prediction Network (BOPN), which predicts the boundary offsets between candidate spans and their nearest entity spans. By leveraging the guiding semantics of boundary offsets, BOPN establishes connections between non-entity and entity spans, enabling non-entity spans to function as additional positive samples for entity detection. Furthermore, our method integrates entity type and span representations to generate type-aware boundary offsets instead of using entity types as detection targets. We conduct experiments on eight widely-used NER datasets, and the results demonstrate that our proposed BOPN outperforms previous state-of-the-art methods.

Kaushik Rangadurai, Siyang Yuan, Minhui Huang et al.

Retrieval, the initial stage of a recommendation system, is tasked with down-selecting items from a pool of tens of millions of candidates to a few thousands. Embedding Based Retrieval (EBR) has been a typical choice for this problem, addressing the computational demands of deep neural networks across vast item corpora. EBR utilizes Two Tower or Siamese Networks to learn representations for users and items, and employ Approximate Nearest Neighbor (ANN) search to efficiently retrieve relevant items. Despite its popularity in industry, EBR faces limitations. The Two Tower architecture, relying on a single dot product interaction, struggles to capture complex data distributions due to limited capability in learning expressive interactions between users and items. Additionally, ANN index building and representation learning for user and item are often separate, leading to inconsistencies exacerbated by representation (e.g. continuous online training) and item drift (e.g. items expired and new items added). In this paper, we introduce the Hierarchical Structured Neural Network (HSNN), an efficient deep neural network model to learn intricate user and item interactions beyond the commonly used dot product in retrieval tasks, achieving sublinear computational costs relative to corpus size. A Modular Neural Network (MoNN) is designed to maintain high expressiveness for interaction learning while ensuring efficiency. A mixture of MoNNs operate on a hierarchical item index to achieve extensive computation sharing, enabling it to scale up to large corpus size. MoNN and the hierarchical index are jointly learnt to continuously adapt to distribution shifts in both user interests and item distributions. HSNN achieves substantial improvement in offline evaluation compared to prevailing methods.

Carolina Zheng, Minhui Huang, Dmitrii Pedchenko et al.

The exponential growth of online content has posed significant challenges to ID-based models in industrial recommendation systems, ranging from extremely high cardinality and dynamically growing ID space, to highly skewed engagement distributions, to prediction instability as a result of natural id life cycles (e.g, the birth of new IDs and retirement of old IDs). To address these issues, many systems rely on random hashing to handle the id space and control the corresponding model parameters (i.e embedding table). However, this approach introduces data pollution from multiple ids sharing the same embedding, leading to degraded model performance and embedding representation instability. This paper examines these challenges and introduces Semantic ID prefix ngram, a novel token parameterization technique that significantly improves the performance of the original Semantic ID. Semantic ID prefix ngram creates semantically meaningful collisions by hierarchically clustering items based on their content embeddings, as opposed to random assignments. Through extensive experimentation, we demonstrate that Semantic ID prefix ngram not only addresses embedding instability but also significantly improves tail id modeling, reduces overfitting, and mitigates representation shifts. We further highlight the advantages of Semantic ID prefix ngram in attention-based models that contextualize user histories, showing substantial performance improvements. We also report our experience of integrating Semantic ID into Meta production Ads Ranking system, leading to notable performance gains and enhanced prediction stability in live deployments.

Faruk Ahmed, Lin Yang, Tiam Jaroensri et al.

The interpretation of histopathology cases underlies many important diagnostic and treatment decisions in medicine. Notably, this process typically requires pathologists to integrate and summarize findings across multiple slides per case. Existing vision-language capabilities in computational pathology have so far been largely limited to small regions of interest, larger regions at low magnification, or single whole-slide images (WSIs). This limits interpretation of findings that span multiple high-magnification regions across multiple WSIs. By making use of Gemini 1.5 Flash, a large multimodal model (LMM) with a 1-million token context window, we demonstrate the ability to generate bottom-line diagnoses from up to 40,000 768x768 pixel image patches from multiple WSIs at 10X magnification. This is the equivalent of up to 11 hours of video at 1 fps. Expert pathologist evaluations demonstrate that the generated report text is clinically accurate and equivalent to or preferred over the original reporting for 68% (95% CI: [60%, 76%]) of multi-slide examples with up to 5 slides. While performance decreased for examples with 6 or more slides, this study demonstrates the promise of leveraging the long-context capabilities of modern LMMs for the uniquely challenging task of medical report generation where each case can contain thousands of image patches.

Hao Yang, Zhiyu Yang, Yunjie Zhang et al.

Chain-of-Thought reasoning has emerged as a pivotal methodology for enhancing model inference capabilities. Despite growing interest in Chain-of-Thought reasoning, its underlying mechanisms remain unclear. This paper explores the working mechanisms of Chain-of-Thought reasoning from the perspective of the dual relationship between in-context learning and pretrained priors. We first conduct a fine-grained lexical-level analysis of rationales to examine the model's reasoning behavior. Then, by incrementally introducing noisy exemplars, we examine how the model balances pretrained priors against erroneous in-context information. Finally, we investigate whether prompt engineering can induce slow thinking in large language models. Our extensive experiments reveal three key findings: (1) The model not only quickly learns the reasoning structure at the lexical level but also grasps deeper logical reasoning patterns, yet it heavily relies on pretrained priors. (2) Providing sufficient exemplars shifts the model's decision-making from pretrained priors to in-context signals, while misleading prompts introduce instability. (3) Long Chain-of-Thought prompting can induce the model to generate longer reasoning chains, thereby improving its performance on downstream tasks.

Yupei Yang, Fan Feng, Lin Yang et al.

Relation extraction enables the construction of structured knowledge for many downstream applications. While large language models (LLMs) have shown great promise in this domain, most existing methods concentrate on relation classification, which predicts the semantic relation type between a related entity pair. However, we observe that LLMs often struggle to reliably determine whether a relation exists, especially in cases involving complex sentence structures or intricate semantics, which leads to spurious predictions. Such hallucinations can introduce noisy edges in knowledge graphs, compromising the integrity of structured knowledge and downstream reliability. To address these challenges, we propose DEPTH, a framework that integrates Dependency-aware sEntence simPlification and Two-tiered Hierarchical refinement into the relation extraction pipeline. Given a sentence and its candidate entity pairs, DEPTH operates in two stages: (1) the Grounding module extracts relations for each pair by leveraging their shortest dependency path, distilling the sentence into a minimal yet coherent relational context that reduces syntactic noise while preserving key semantics; (2) the Refinement module aggregates all local predictions and revises them based on a holistic understanding of the sentence, correcting omissions and inconsistencies. We further introduce a causality-driven reward model that mitigates reward hacking by disentangling spurious correlations, enabling robust fine-tuning via reinforcement learning with human feedback. Experiments on six benchmarks demonstrate that DEPTH reduces the average hallucination rate to 7.0\% while achieving a 17.2\% improvement in average F1 score over state-of-the-art baselines.

Hao Zheng, Jun Han, Hongxiao Wang et al.

A large labeled dataset is a key to the success of supervised deep learning, but for medical image segmentation, it is highly challenging to obtain sufficient annotated images for model training. In many scenarios, unannotated images are abundant and easy to acquire. Self-supervised learning (SSL) has shown great potentials in exploiting raw data information and representation learning. In this paper, we propose Hierarchical Self-Supervised Learning (HSSL), a new self-supervised framework that boosts medical image segmentation by making good use of unannotated data. Unlike the current literature on task-specific self-supervised pretraining followed by supervised fine-tuning, we utilize SSL to learn task-agnostic knowledge from heterogeneous data for various medical image segmentation tasks. Specifically, we first aggregate a dataset from several medical challenges, then pre-train the network in a self-supervised manner, and finally fine-tune on labeled data. We develop a new loss function by combining contrastive loss and classification loss and pretrain an encoder-decoder architecture for segmentation tasks. Our extensive experiments show that multi-domain joint pre-training benefits downstream segmentation tasks and outperforms single-domain pre-training significantly. Compared to learning from scratch, our new method yields better performance on various tasks (e.g., +0.69% to +18.60% in Dice scores with 5% of annotated data). With limited amounts of training data, our method can substantially bridge the performance gap w.r.t. denser annotations (e.g., 10% vs.~100% of annotated data).

Jiatong Cai, Chenglu Zhu, Can Cui et al.

Ki67 is a significant biomarker in the diagnosis and prognosis of cancer, whose index can be evaluated by quantifying its expression in Ki67 immunohistochemistry (IHC) stained images. However, quantitative analysis on multi-source Ki67 images is yet a challenging task in practice due to cross-domain distribution differences, which result from imaging variation, staining styles, and lesion types. Many recent studies have made some efforts on domain generalization (DG), whereas there are still some noteworthy limitations. Specifically in the case of Ki67 images, learning invariant representation is at the mercy of the insufficient number of domains and the cell categories mismatching in different domains. In this paper, we propose a novel method to improve DG by searching the domain-agnostic subnetwork in a domain merging scenario. Partial model parameters are iteratively pruned according to the domain gap, which is caused by the data converting from a single domain into merged domains during training. In addition, the model is optimized by fine-tuning on merged domains to eliminate the interference of class mismatching among various domains. Furthermore, an appropriate implementation is attained by applying the pruning method to different parts of the framework. Compared with known DG methods, our method yields excellent performance in multiclass nucleus recognition of Ki67 IHC images, especially in the lost category cases. Moreover, our competitive results are also evaluated on the public dataset over the state-of-the-art DG methods.

Hongxiao Wang, Hao Zheng, Jianxu Chen et al.

Automatic histopathology image segmentation is crucial to disease analysis. Limited available labeled data hinders the generalizability of trained models under the fully supervised setting. Semi-supervised learning (SSL) based on generative methods has been proven to be effective in utilizing diverse image characteristics. However, it has not been well explored what kinds of generated images would be more useful for model training and how to use such images. In this paper, we propose a new data guided generative method for histopathology image segmentation by leveraging the unlabeled data distributions. First, we design an image generation module. Image content and style are disentangled and embedded in a clustering-friendly space to utilize their distributions. New images are synthesized by sampling and cross-combining contents and styles. Second, we devise an effective data selection policy for judiciously sampling the generated images: (1) to make the generated training set better cover the dataset, the clusters that are underrepresented in the original training set are covered more; (2) to make the training process more effective, we identify and oversample the images of "hard cases" in the data for which annotated training data may be scarce. Our method is evaluated on glands and nuclei datasets. We show that under both the inductive and transductive settings, our SSL method consistently boosts the performance of common segmentation models and attains state-of-the-art results.

Qi She, Fan Feng, Qi Liu et al.

This report summarizes IROS 2019-Lifelong Robotic Vision Competition (Lifelong Object Recognition Challenge) with methods and results from the top $8$ finalists (out of over~$150$ teams). The competition dataset (L)ifel(O)ng (R)obotic V(IS)ion (OpenLORIS) - Object Recognition (OpenLORIS-object) is designed for driving lifelong/continual learning research and application in robotic vision domain, with everyday objects in home, office, campus, and mall scenarios. The dataset explicitly quantifies the variants of illumination, object occlusion, object size, camera-object distance/angles, and clutter information. Rules are designed to quantify the learning capability of the robotic vision system when faced with the objects appearing in the dynamic environments in the contest. Individual reports, dataset information, rules, and released source code can be found at the project homepage: "https://lifelong-robotic-vision.github.io/competition/".

Baohua Sun, Lin Yang, Hao Sha et al.

Recent years NLP research has witnessed the record-breaking accuracy improvement by DNN models. However, power consumption is one of the practical concerns for deploying NLP systems. Most of the current state-of-the-art algorithms are implemented on GPUs, which is not power-efficient and the deployment cost is also very high. On the other hand, CNN Domain Specific Accelerator (CNN-DSA) has been in mass production providing low-power and low cost computation power. In this paper, we will implement the Super Characters method on the CNN-DSA. In addition, we modify the Super Characters method to utilize the multi-modal data, i.e. text plus tabular data in the CL-Aff sharedtask.

Baohua Sun, Lin Yang, Michael Lin et al.

Power-efficient CNN Domain Specific Accelerator (CNN-DSA) chips are currently available for wide use in mobile devices. These chips are mainly used in computer vision applications. However, the recent work of Super Characters method for text classification and sentiment analysis tasks using two-dimensional CNN models has also achieved state-of-the-art results through the method of transfer learning from vision to text. In this paper, we implemented the text classification and sentiment analysis applications on mobile devices using CNN-DSA chips. Compact network representations using one-bit and three-bits precision for coefficients and five-bits for activations are used in the CNN-DSA chip with power consumption less than 300mW. For edge devices under memory and compute constraints, the network is further compressed by approximating the external Fully Connected (FC) layers within the CNN-DSA chip. At the workshop, we have two system demonstrations for NLP tasks. The first demo classifies the input English Wikipedia sentence into one of the 14 ontologies. The second demo classifies the Chinese online-shopping review into positive or negative.

Baohua Sun, Lin Yang, Michael Lin et al.

Language and vision are processed as two different modal in current work for image captioning. However, recent work on Super Characters method shows the effectiveness of two-dimensional word embedding, which converts text classification problem into image classification problem. In this paper, we propose the SuperCaptioning method, which borrows the idea of two-dimensional word embedding from Super Characters method, and processes the information of language and vision together in one single CNN model. The experimental results on Flickr30k data shows the proposed method gives high quality image captions. An interactive demo is ready to show at the workshop.

Baohua Sun, Lin Yang, Michael Lin et al.

The recent work of Super Characters method using two-dimensional word embedding achieved state-of-the-art results in text classification tasks, showcasing the promise of this new approach. This paper borrows the idea of Super Characters method and two-dimensional embedding, and proposes a method of generating conversational response for open domain dialogues. The experimental results on a public dataset shows that the proposed SuperChat method generates high quality responses. An interactive demo is ready to show at the workshop.

Baohua Sun, Lin Yang, Wenhan Zhang et al.

Tabular data is the most commonly used form of data in industry. Gradient Boosting Trees, Support Vector Machine, Random Forest, and Logistic Regression are typically used for classification tasks on tabular data. DNN models using categorical embeddings are also applied in this task, but all attempts thus far have used one-dimensional embeddings. The recent work of Super Characters method using two-dimensional word embeddings achieved the state of art result in text classification tasks, showcasing the promise of this new approach. In this paper, we propose the SuperTML method, which borrows the idea of Super Characters method and two-dimensional embeddings to address the problem of classification on tabular data. For each input of tabular data, the features are first projected into two-dimensional embeddings like an image, and then this image is fed into fine-tuned two-dimensional CNN models for classification. Experimental results have shown that the proposed SuperTML method had achieved state-of-the-art results on both large and small datasets.

Baohua Sun, Lin Yang, Catherine Chi et al.

The Super Characters method addresses sentiment analysis problems by first converting the input text into images and then applying 2D-CNN models to classify the sentiment. It achieves state of the art performance on many benchmark datasets. However, it is not as straightforward to apply in Latin languages as in Asian languages. Because the 2D-CNN model is designed to recognize two-dimensional images, it is better if the inputs are in the form of glyphs. In this paper, we propose SEW (Squared English Word) method generating a squared glyph for each English word by drawing Super Characters images of each English word at the alphabet level, combining the squared glyph together into a whole Super Characters image at the sentence level, and then applying the CNN model to classify the sentiment within the sentence. We applied the SEW method to Wikipedia dataset and obtained a 2.1% accuracy gain compared to the original Super Characters method. For multi-modal data with both structured tabular data and unstructured natural language text, the modified SEW method integrates the data into a single image and classifies sentiment with one unified CNN model.

Peixian Liang, Jianxu Chen, Hao Zheng et al.

The Encoder-Decoder architecture is a main stream deep learning model for biomedical image segmentation. The encoder fully compresses the input and generates encoded features, and the decoder then produces dense predictions using encoded features. However, decoders are still under-explored in such architectures. In this paper, we comprehensively study the state-of-the-art Encoder-Decoder architectures, and propose a new universal decoder, called cascade decoder, to improve semantic segmentation accuracy. Our cascade decoder can be embedded into existing networks and trained altogether in an end-to-end fashion. The cascade decoder structure aims to conduct more effective decoding of hierarchically encoded features and is more compatible with common encoders than the known decoders. We replace the decoders of state-of-the-art models with our cascade decoder for several challenging biomedical image segmentation tasks, and the considerable improvements achieved demonstrate the efficacy of our new decoding method.

Hao Zheng, Yizhe Zhang, Lin Yang et al.

3D image segmentation plays an important role in biomedical image analysis. Many 2D and 3D deep learning models have achieved state-of-the-art segmentation performance on 3D biomedical image datasets. Yet, 2D and 3D models have their own strengths and weaknesses, and by unifying them together, one may be able to achieve more accurate results. In this paper, we propose a new ensemble learning framework for 3D biomedical image segmentation that combines the merits of 2D and 3D models. First, we develop a fully convolutional network based meta-learner to learn how to improve the results from 2D and 3D models (base-learners). Then, to minimize over-fitting for our sophisticated meta-learner, we devise a new training method that uses the results of the base-learners as multiple versions of "ground truths". Furthermore, since our new meta-learner training scheme does not depend on manual annotation, it can utilize abundant unlabeled 3D image data to further improve the model. Extensive experiments on two public datasets (the HVSMR 2016 Challenge dataset and the mouse piriform cortex dataset) show that our approach is effective under fully-supervised, semi-supervised, and transductive settings, and attains superior performance over state-of-the-art image segmentation methods.

Xiaoshuang Shi, Fuyong Xing, Zizhao Zhang et al.

Maintaining the pair similarity relationship among originally high-dimensional data into a low-dimensional binary space is a popular strategy to learn binary codes. One simiple and intutive method is to utilize two identical code matrices produced by hash functions to approximate a pairwise real label matrix. However, the resulting quartic problem is difficult to directly solve due to the non-convex and non-smooth nature of the objective. In this paper, unlike previous optimization methods using various relaxation strategies, we aim to directly solve the original quartic problem using a novel alternative optimization mechanism to linearize the quartic problem by introducing a linear regression model. Additionally, we find that gradually learning each batch of binary codes in a sequential mode, i.e. batch by batch, is greatly beneficial to the convergence of binary code learning. Based on this significant discovery and the proposed strategy, we introduce a scalable symmetric discrete hashing algorithm that gradually and smoothly updates each batch of binary codes. To further improve the smoothness, we also propose a greedy symmetric discrete hashing algorithm to update each bit of batch binary codes. Moreover, we extend the proposed optimization mechanism to solve the non-convex optimization problems for binary code learning in many other pairwise based hashing algorithms. Extensive experiments on benchmark single-label and multi-label databases demonstrate the superior performance of the proposed mechanism over recent state-of-the-art methods.

Baohua Sun, Lin Yang, Patrick Dong et al.

We propose a method named Super Characters for sentiment classification. This method converts the sentiment classification problem into image classification problem by projecting texts into images and then applying CNN models for classification. Text features are extracted automatically from the generated Super Characters images, hence there is no need of any explicit step of embedding the words or characters into numerical vector representations. Experimental results on large social media corpus show that the Super Characters method consistently outperforms other methods for sentiment classification and topic classification tasks on ten large social media datasets of millions of contents in four different languages, including Chinese, Japanese, Korean and English.

Jinzheng Cai, Youbao Tang, Le Lu et al.

Volumetric lesion segmentation from computed tomography (CT) images is a powerful means to precisely assess multiple time-point lesion/tumor changes. However, because manual 3D segmentation is prohibitively time consuming, current practices rely on an imprecise surrogate called response evaluation criteria in solid tumors (RECIST). Despite their coarseness, RECIST markers are commonly found in current hospital picture and archiving systems (PACS), meaning they can provide a potentially powerful, yet extraordinarily challenging, source of weak supervision for full 3D segmentation. Toward this end, we introduce a convolutional neural network (CNN) based weakly supervised slice-propagated segmentation (WSSS) method to 1) generate the initial lesion segmentation on the axial RECIST-slice; 2) learn the data distribution on RECIST-slices; 3) extrapolate to segment the whole lesion slice by slice to finally obtain a volumetric segmentation. To validate the proposed method, we first test its performance on a fully annotated lymph node dataset, where WSSS performs comparably to its fully supervised counterparts. We then test on a comprehensive lesion dataset with 32,735 RECIST marks, where we report a mean Dice score of 92% on RECIST-marked slices and 76% on the entire 3D volumes.

Zhuo Zhao, Lin Yang, Hao Zheng et al.

Instance segmentation in 3D images is a fundamental task in biomedical image analysis. While deep learning models often work well for 2D instance segmentation, 3D instance segmentation still faces critical challenges, such as insufficient training data due to various annotation difficulties in 3D biomedical images. Common 3D annotation methods (e.g., full voxel annotation) incur high workloads and costs for labeling enough instances for training deep learning 3D instance segmentation models. In this paper, we propose a new weak annotation approach for training a fast deep learning 3D instance segmentation model without using full voxel mask annotation. Our approach needs only 3D bounding boxes for all instances and full voxel annotation for a small fraction of the instances, and uses a novel two-stage 3D instance segmentation model utilizing these two kinds of annotation, respectively. We evaluate our approach on several biomedical image datasets, and the experimental results show that (1) with full annotated boxes and a small amount of masks, our approach can achieve similar performance as the best known methods using full annotation, and (2) with similar annotation time, our approach outperforms the best known methods that use full annotation.

Lin Yang, Yizhe Zhang, Zhuo Zhao et al.

In recent years, deep learning (DL) methods have become powerful tools for biomedical image segmentation. However, high annotation efforts and costs are commonly needed to acquire sufficient biomedical training data for DL models. To alleviate the burden of manual annotation, in this paper, we propose a new weakly supervised DL approach for biomedical image segmentation using boxes only annotation. First, we develop a method to combine graph search (GS) and DL to generate fine object masks from box annotation, in which DL uses box annotation to compute a rough segmentation for GS and then GS is applied to locate the optimal object boundaries. During the mask generation process, we carefully utilize information from box annotation to filter out potential errors, and then use the generated masks to train an accurate DL segmentation network. Extensive experiments on gland segmentation in histology images, lymph node segmentation in ultrasound images, and fungus segmentation in electron microscopy images show that our approach attains superior performance over the best known state-of-the-art weakly supervised DL method and is able to achieve (1) nearly the same accuracy compared to fully supervised DL methods with far less annotation effort, (2) significantly better results with similar annotation time, and (3) robust performance in various applications.

Baohua Sun, Lin Yang, Patrick Dong et al.

Computer vision performances have been significantly improved in recent years by Convolutional Neural Networks(CNN). Currently, applications using CNN algorithms are deployed mainly on general purpose hardwares, such as CPUs, GPUs or FPGAs. However, power consumption, speed, accuracy, memory footprint, and die size should all be taken into consideration for mobile and embedded applications. Domain Specific Architecture (DSA) for CNN is the efficient and practical solution for CNN deployment and implementation. We designed and produced a 28nm Two-Dimensional CNN-DSA accelerator with an ultra power-efficient performance of 9.3TOPS/Watt and with all processing done in the internal memory instead of outside DRAM. It classifies 224x224 RGB image inputs at more than 140fps with peak power consumption at less than 300mW and an accuracy comparable to the VGG benchmark. The CNN-DSA accelerator is reconfigurable to support CNN model coefficients of various layer sizes and layer types, including convolution, depth-wise convolution, short-cut connections, max pooling, and ReLU. Furthermore, in order to better support real-world deployment for various application scenarios, especially with low-end mobile and embedded platforms and MCUs (Microcontroller Units), we also designed algorithms to fully utilize the CNN-DSA accelerator efficiently by reducing the dependency on external accelerator computation resources, including implementation of Fully-Connected (FC) layers within the accelerator and compression of extracted features from the CNN-DSA accelerator. Live demos with our CNN-DSA accelerator on mobile and embedded systems show its capabilities to be widely and practically applied in the real world.

Jinzheng Cai, Le Lu, Fuyong Xing et al.

Automatic pancreas segmentation in radiology images, eg., computed tomography (CT) and magnetic resonance imaging (MRI), is frequently required by computer-aided screening, diagnosis, and quantitative assessment. Yet pancreas is a challenging abdominal organ to segment due to the high inter-patient anatomical variability in both shape and volume metrics. Recently, convolutional neural networks (CNNs) have demonstrated promising performance on accurate segmentation of pancreas. However, the CNN-based method often suffers from segmentation discontinuity for reasons such as noisy image quality and blurry pancreatic boundary. From this point, we propose to introduce recurrent neural networks (RNNs) to address the problem of spatial non-smoothness of inter-slice pancreas segmentation across adjacent image slices. To inference initial segmentation, we first train a 2D CNN sub-network, where we modify its network architecture with deep-supervision and multi-scale feature map aggregation so that it can be trained from scratch with small-sized training data and presents superior performance than transferred models. Thereafter, the successive CNN outputs are processed by another RNN sub-network, which refines the consistency of segmented shapes. More specifically, the RNN sub-network consists convolutional long short-term memory (CLSTM) units in both top-down and bottom-up directions, which regularizes the segmentation of an image by integrating predictions of its neighboring slices. We train the stacked CNN-RNN model end-to-end and perform quantitative evaluations on both CT and MRI images.

Xiaowei Xu, Qing Lu, Yu Hu et al.

With pervasive applications of medical imaging in health-care, biomedical image segmentation plays a central role in quantitative analysis, clinical diagno- sis, and medical intervention. Since manual anno- tation su ers limited reproducibility, arduous e orts, and excessive time, automatic segmentation is desired to process increasingly larger scale histopathological data. Recently, deep neural networks (DNNs), par- ticularly fully convolutional networks (FCNs), have been widely applied to biomedical image segmenta- tion, attaining much improved performance. At the same time, quantization of DNNs has become an ac- tive research topic, which aims to represent weights with less memory (precision) to considerably reduce memory and computation requirements of DNNs while maintaining acceptable accuracy. In this paper, we apply quantization techniques to FCNs for accurate biomedical image segmentation. Unlike existing litera- ture on quantization which primarily targets memory and computation complexity reduction, we apply quan- tization as a method to reduce over tting in FCNs for better accuracy. Speci cally, we focus on a state-of- the-art segmentation framework, suggestive annotation [22], which judiciously extracts representative annota- tion samples from the original training dataset, obtain- ing an e ective small-sized balanced training dataset. We develop two new quantization processes for this framework: (1) suggestive annotation with quantiza- tion for highly representative training samples, and (2) network training with quantization for high accuracy. Extensive experiments on the MICCAI Gland dataset show that both quantization processes can improve the segmentation performance, and our proposed method exceeds the current state-of-the-art performance by up to 1%. In addition, our method has a reduction of up to 6.4x on memory usage.

Peixian Liang, Jianxu Chen, Pavel A. Brodskiy et al.

Wing disc pouches of fruit flies are a powerful genetic model for studying physiological intercellular calcium ($Ca^{2+}$) signals for dynamic analysis of cell signaling in organ development and disease studies. A key to analyzing spatial-temporal patterns of $Ca^{2+}$ signal waves is to accurately align the pouches across image sequences. However, pouches in different image frames may exhibit extensive intensity oscillations due to $Ca^{2+}$ signaling dynamics, and commonly used multimodal non-rigid registration methods may fail to achieve satisfactory results. In this paper, we develop a new two-phase non-rigid registration approach to register pouches in image sequences. First, we conduct segmentation of the region of interest. (i.e., pouches) using a deep neural network model. Second, we obtain an optimal transformation and align pouches across the image sequences. Evaluated using both synthetic data and real pouch data, our method considerably outperforms the state-of-the-art non-rigid registration methods.

Jinzheng Cai, Youbao Tang, Le Lu et al.

Volumetric lesion segmentation via medical imaging is a powerful means to precisely assess multiple time-point lesion/tumor changes. Because manual 3D segmentation is prohibitively time consuming and requires radiological experience, current practices rely on an imprecise surrogate called response evaluation criteria in solid tumors (RECIST). Despite their coarseness, RECIST marks are commonly found in current hospital picture and archiving systems (PACS), meaning they can provide a potentially powerful, yet extraordinarily challenging, source of weak supervision for full 3D segmentation. Toward this end, we introduce a convolutional neural network based weakly supervised self-paced segmentation (WSSS) method to 1) generate the initial lesion segmentation on the axial RECIST-slice; 2) learn the data distribution on RECIST-slices; 3) adapt to segment the whole volume slice by slice to finally obtain a volumetric segmentation. In addition, we explore how super-resolution images (2~5 times beyond the physical CT imaging), generated from a proposed stacked generative adversarial network, can aid the WSSS performance. We employ the DeepLesion dataset, a comprehensive CT-image lesion dataset of 32,735 PACS-bookmarked findings, which include lesions, tumors, and lymph nodes of varying sizes, categories, body regions and surrounding contexts. These are drawn from 10,594 studies of 4,459 patients. We also validate on a lymph-node dataset, where 3D ground truth masks are available for all images. For the DeepLesion dataset, we report mean Dice coefficients of 93% on RECIST-slices and 76% in 3D lesion volumes. We further validate using a subjective user study, where an experienced radiologist accepted our WSSS-generated lesion segmentation results with a high probability of 92.4%.

Zizhao Zhang, Fuyong Xing, Hai Su et al.

The fast growing deep learning technologies have become the main solution of many machine learning problems for medical image analysis. Deep convolution neural networks (CNNs), as one of the most important branch of the deep learning family, have been widely investigated for various computer-aided diagnosis tasks including long-term problems and continuously emerging new problems. Image contour detection is a fundamental but challenging task that has been studied for more than four decades. Recently, we have witnessed the significantly improved performance of contour detection thanks to the development of CNNs. Beyond purusing performance in existing natural image benchmarks, contour detection plays a particularly important role in medical image analysis. Segmenting various objects from radiology images or pathology images requires accurate detection of contours. However, some problems, such as discontinuity and shape constraints, are insufficiently studied in CNNs. It is necessary to clarify the challenges to encourage further exploration. The performance of CNN based contour detection relies on the state-of-the-art CNN architectures. Careful investigation of their design principles and motivations is critical and beneficial to contour detection. In this paper, we first review recent development of medical image contour detection and point out the current confronting challenges and problems. We discuss the development of general CNNs and their applications in image contours (or edges) detection. We compare those methods in detail, clarify their strengthens and weaknesses. Then we review their recent applications in medical image analysis and point out limitations, with the goal to light some potential directions in medical image analysis. We expect the paper to cover comprehensive technical ingredients of advanced CNNs to enrich the study in the medical image domain.

Jinzheng Cai, Le Lu, Yuanpu Xie et al.

Deep neural networks have demonstrated very promising performance on accurate segmentation of challenging organs (e.g., pancreas) in abdominal CT and MRI scans. The current deep learning approaches conduct pancreas segmentation by processing sequences of 2D image slices independently through deep, dense per-pixel masking for each image, without explicitly enforcing spatial consistency constraint on segmentation of successive slices. We propose a new convolutional/recurrent neural network architecture to address the contextual learning and segmentation consistency problem. A deep convolutional sub-network is first designed and pre-trained from scratch. The output layer of this network module is then connected to recurrent layers and can be fine-tuned for contextual learning, in an end-to-end manner. Our recurrent sub-network is a type of Long short-term memory (LSTM) network that performs segmentation on an image by integrating its neighboring slice segmentation predictions, in the form of a dependent sequence processing. Additionally, a novel segmentation-direct loss function (named Jaccard Loss) is proposed and deep networks are trained to optimize Jaccard Index (JI) directly. Extensive experiments are conducted to validate our proposed deep models, on quantitative pancreas segmentation using both CT and MRI scans. Our method outperforms the state-of-the-art work on CT [11] and MRI pancreas segmentation [1], respectively.

Lin Yang, Yizhe Zhang, Jianxu Chen et al.

Image segmentation is a fundamental problem in biomedical image analysis. Recent advances in deep learning have achieved promising results on many biomedical image segmentation benchmarks. However, due to large variations in biomedical images (different modalities, image settings, objects, noise, etc), to utilize deep learning on a new application, it usually needs a new set of training data. This can incur a great deal of annotation effort and cost, because only biomedical experts can annotate effectively, and often there are too many instances in images (e.g., cells) to annotate. In this paper, we aim to address the following question: With limited effort (e.g., time) for annotation, what instances should be annotated in order to attain the best performance? We present a deep active learning framework that combines fully convolutional network (FCN) and active learning to significantly reduce annotation effort by making judicious suggestions on the most effective annotation areas. We utilize uncertainty and similarity information provided by FCN and formulate a generalized version of the maximum set cover problem to determine the most representative and uncertain areas for annotation. Extensive experiments using the 2015 MICCAI Gland Challenge dataset and a lymph node ultrasound image segmentation dataset show that, using annotation suggestions by our method, state-of-the-art segmentation performance can be achieved by using only 50% of training data.

Jianxu Chen, Lin Yang, Yizhe Zhang et al.

Segmentation of 3D images is a fundamental problem in biomedical image analysis. Deep learning (DL) approaches have achieved state-of-the-art segmentation perfor- mance. To exploit the 3D contexts using neural networks, known DL segmentation methods, including 3D convolution, 2D convolution on planes orthogonal to 2D image slices, and LSTM in multiple directions, all suffer incompatibility with the highly anisotropic dimensions in common 3D biomedical images. In this paper, we propose a new DL framework for 3D image segmentation, based on a com- bination of a fully convolutional network (FCN) and a recurrent neural network (RNN), which are responsible for exploiting the intra-slice and inter-slice contexts, respectively. To our best knowledge, this is the first DL framework for 3D image segmentation that explicitly leverages 3D image anisotropism. Evaluating using a dataset from the ISBI Neuronal Structure Segmentation Challenge and in-house image stacks for 3D fungus segmentation, our approach achieves promising results comparing to the known DL-based 3D segmentation approaches.

Changsheng Li, Qingshan Liu, Weishan Dong et al.

In this paper, we propose a new max-margin based discriminative feature learning method. Specifically, we aim at learning a low-dimensional feature representation, so as to maximize the global margin of the data and make the samples from the same class as close as possible. In order to enhance the robustness to noise, a $l_{2,1}$ norm constraint is introduced to make the transformation matrix in group sparsity. In addition, for multi-class classification tasks, we further intend to learn and leverage the correlation relationships among multiple class tasks for assisting in learning discriminative features. The experimental results demonstrate the power of the proposed method against the related state-of-the-art methods.