Seyed-Ahmad Ahmadi

Jianning Li, Zongwei Zhou, Jiancheng Yang et al.

Prior to the deep learning era, shape was commonly used to describe the objects. Nowadays, state-of-the-art (SOTA) algorithms in medical imaging are predominantly diverging from computer vision, where voxel grids, meshes, point clouds, and implicit surface models are used. This is seen from numerous shape-related publications in premier vision conferences as well as the growing popularity of ShapeNet (about 51,300 models) and Princeton ModelNet (127,915 models). For the medical domain, we present a large collection of anatomical shapes (e.g., bones, organs, vessels) and 3D models of surgical instrument, called MedShapeNet, created to facilitate the translation of data-driven vision algorithms to medical applications and to adapt SOTA vision algorithms to medical problems. As a unique feature, we directly model the majority of shapes on the imaging data of real patients. As of today, MedShapeNet includes 23 dataset with more than 100,000 shapes that are paired with annotations (ground truth). Our data is freely accessible via a web interface and a Python application programming interface (API) and can be used for discriminative, reconstructive, and variational benchmarks as well as various applications in virtual, augmented, or mixed reality, and 3D printing. Exemplary, we present use cases in the fields of classification of brain tumors, facial and skull reconstructions, multi-class anatomy completion, education, and 3D printing. In future, we will extend the data and improve the interfaces. The project pages are: https://medshapenet.ikim.nrw/ and https://github.com/Jianningli/medshapenet-feedback

Jianning Li, Jana Fragemann, Seyed-Ahmad Ahmadi et al.

The reconstruction loss and the Kullback-Leibler divergence (KLD) loss in a variational autoencoder (VAE) often play antagonistic roles, and tuning the weight of the KLD loss in $β$-VAE to achieve a balance between the two losses is a tricky and dataset-specific task. As a result, current practices in VAE training often result in a trade-off between the reconstruction fidelity and the continuity$/$disentanglement of the latent space, if the weight $β$ is not carefully tuned. In this paper, we present intuitions and a careful analysis of the antagonistic mechanism of the two losses, and propose, based on the insights, a simple yet effective two-stage method for training a VAE. Specifically, the method aggregates a learned Gaussian posterior $z \sim q_θ (z|x)$ with a decoder decoupled from the KLD loss, which is trained to learn a new conditional distribution $p_φ (x|z)$ of the input data $x$. Experimentally, we show that the aggregated VAE maximally satisfies the Gaussian assumption about the latent space, while still achieves a reconstruction error comparable to when the latent space is only loosely regularized by $\mathcal{N}(\mathbf{0},I)$. The proposed approach does not require hyperparameter (i.e., the KLD weight $β$) tuning given a specific dataset as required in common VAE training practices. We evaluate the method using a medical dataset intended for 3D skull reconstruction and shape completion, and the results indicate promising generative capabilities of the VAE trained using the proposed method. Besides, through guided manipulation of the latent variables, we establish a connection between existing autoencoder (AE)-based approaches and generative approaches, such as VAE, for the shape completion problem. Codes and pre-trained weights are available at https://github.com/Jianningli/skullVAE

Jianning Li, André Ferreira, Behrus Puladi et al.

We present a deep learning-based approach for skull reconstruction for MONAI, which has been pre-trained on the MUG500+ skull dataset. The implementation follows the MONAI contribution guidelines, hence, it can be easily tried out and used, and extended by MONAI users. The primary goal of this paper lies in the investigation of open-sourcing codes and pre-trained deep learning models under the MONAI framework. Nowadays, open-sourcing software, especially (pre-trained) deep learning models, has become increasingly important. Over the years, medical image analysis experienced a tremendous transformation. Over a decade ago, algorithms had to be implemented and optimized with low-level programming languages, like C or C++, to run in a reasonable time on a desktop PC, which was not as powerful as today's computers. Nowadays, users have high-level scripting languages like Python, and frameworks like PyTorch and TensorFlow, along with a sea of public code repositories at hand. As a result, implementations that had thousands of lines of C or C++ code in the past, can now be scripted with a few lines and in addition executed in a fraction of the time. To put this even on a higher level, the Medical Open Network for Artificial Intelligence (MONAI) framework tailors medical imaging research to an even more convenient process, which can boost and push the whole field. The MONAI framework is a freely available, community-supported, open-source and PyTorch-based framework, that also enables to provide research contributions with pre-trained models to others. Codes and pre-trained weights for skull reconstruction are publicly available at: https://github.com/Project-MONAI/research-contributions/tree/master/SkullRec

Kamilia Mullakaeva, Luca Cosmo, Anees Kazi et al.

Graphs are a powerful tool for representing and analyzing unstructured, non-Euclidean data ubiquitous in the healthcare domain. Two prominent examples are molecule property prediction and brain connectome analysis. Importantly, recent works have shown that considering relationships between input data samples have a positive regularizing effect for the downstream task in healthcare applications. These relationships are naturally modeled by a (possibly unknown) graph structure between input samples. In this work, we propose Graph-in-Graph (GiG), a neural network architecture for protein classification and brain imaging applications that exploits the graph representation of the input data samples and their latent relation. We assume an initially unknown latent-graph structure between graph-valued input data and propose to learn end-to-end a parametric model for message passing within and across input graph samples, along with the latent structure connecting the input graphs. Further, we introduce a degree distribution loss that helps regularize the predicted latent relationships structure. This regularization can significantly improve the downstream task. Moreover, the obtained latent graph can represent patient population models or networks of molecule clusters, providing a level of interpretability and knowledge discovery in the input domain of particular value in healthcare.

Yousef Yeganeh, Azade Farshad, Peter Weinberger et al.

Although purely transformer-based architectures showed promising performance in many computer vision tasks, many hybrid models consisting of CNN and transformer blocks are introduced to fit more specialized tasks. Nevertheless, despite the performance gain of both pure and hybrid transformer-based architectures compared to CNNs in medical imaging segmentation, their high training cost and complexity make it challenging to use them in real scenarios. In this work, we propose simple architectures based on purely convolutional layers, and show that by just taking advantage of the attention map visualizations obtained from a self-supervised pretrained vision transformer network (e.g., DINO) one can outperform complex transformer-based networks with much less computation costs. The proposed architecture is composed of two encoder branches with the original image as input in one branch and the attention map visualizations of the same image from multiple self-attention heads from a pre-trained DINO model (as multiple channels) in the other branch. The results of our experiments on two publicly available medical imaging datasets show that the proposed pipeline outperforms U-Net and the state-of-the-art medical image segmentation models.

Fausto Milletari, Johann Frei, Seyed-Ahmad Ahmadi

Deep learning has been recently applied to a multitude of computer vision and medical image analysis problems. Although recent research efforts have improved the state of the art, most of the methods cannot be easily accessed, compared or used by either researchers or the general public. Researchers often publish their code and trained models on the internet, but this does not always enable these approaches to be easily used or integrated in stand-alone applications and existing workflows. In this paper we propose a framework which allows easy deployment and access of deep learning methods for segmentation through a cloud-based architecture. Our approach comprises three parts: a server, which wraps trained deep learning models and their pre- and post-processing data pipelines and makes them available on the cloud; a client which interfaces with the server to obtain predictions on user data; a service registry that informs clients about available prediction endpoints that are available in the cloud. These three parts constitute the open-source TOMAAT framework.

Lasse Hansen, Wiebke Heyer, Christoph Großbröhmer et al.

Medical image registration is critical for clinical applications, and fair benchmarking of different methods is essential for monitoring ongoing progress. To date, the Learn2Reg 2020-2023 challenges have released several complementary datasets and established metrics for evaluations. However, these editions did not capture all aspects of the registration problem, particularly in terms of modality diversity and task complexity. To address these limitations, the 2024 edition introduces three new tasks, including large-scale multi-modal registration and unsupervised inter-subject brain registration, as well as the first microscopy-focused benchmark within Learn2Reg. The new datasets also inspired new method developments, including invertibility constraints, pyramid features, keypoints alignment and instance optimisation.

Gerome Vivar, Anees Kazi, Hendrik Burwinkel et al.

Large-scale population-based studies in medicine are a key resource towards better diagnosis, monitoring, and treatment of diseases. They also serve as enablers of clinical decision support systems, in particular Computer Aided Diagnosis (CADx) using machine learning (ML). Numerous ML approaches for CADx have been proposed in literature. However, these approaches assume full data availability, which is not always feasible in clinical data. To account for missing data, incomplete data samples are either removed or imputed, which could lead to data bias and may negatively affect classification performance. As a solution, we propose an end-to-end learning of imputation and disease prediction of incomplete medical datasets via Multigraph Geometric Matrix Completion (MGMC). MGMC uses multiple recurrent graph convolutional networks, where each graph represents an independent population model based on a key clinical meta-feature like age, sex, or cognitive function. Graph signal aggregation from local patient neighborhoods, combined with multigraph signal fusion via self-attention, has a regularizing effect on both matrix reconstruction and classification performance. Our proposed approach is able to impute class relevant features as well as perform accurate classification on two publicly available medical datasets. We empirically show the superiority of our proposed approach in terms of classification and imputation performance when compared with state-of-the-art approaches. MGMC enables disease prediction in multimodal and incomplete medical datasets. These findings could serve as baseline for future CADx approaches which utilize incomplete datasets.

Hendrik Burwinkel, Holger Matz, Stefan Saur et al.

Today, cataract surgery is the most frequently performed ophthalmic surgery in the world. The cataract, a developing opacity of the human eye lens, constitutes the world's most frequent cause for blindness. During surgery, the lens is removed and replaced by an artificial intraocular lens (IOL). To prevent patients from needing strong visual aids after surgery, a precise prediction of the optical properties of the inserted IOL is crucial. There has been lots of activity towards developing methods to predict these properties from biometric eye data obtained by OCT devices, recently also by employing machine learning. They consider either only biometric data or physical models, but rarely both, and often neglect the IOL geometry. In this work, we propose OpticNet, a novel optical refraction network, loss function, and training scheme which is unsupervised, domain-specific, and physically motivated. We derive a precise light propagation eye model using single-ray raytracing and formulate a differentiable loss function that back-propagates physical gradients into the network. Further, we propose a new transfer learning procedure, which allows unsupervised training on the physical model and fine-tuning of the network on a cohort of real IOL patient cases. We show that our network is not only superior to systems trained with standard procedures but also that our method outperforms the current state of the art in IOL calculation when compared on two biometric data sets.

Hendrik Burwinkel, Matthias Keicher, David Bani-Harouni et al.

Every day, poison control centers (PCC) are called for immediate classification and treatment recommendations if an acute intoxication is suspected. Due to the time-sensitive nature of these cases, doctors are required to propose a correct diagnosis and intervention within a minimal time frame. Usually the toxin is known and recommendations can be made accordingly. However, in challenging cases only symptoms are mentioned and doctors have to rely on their clinical experience. Medical experts and our analyses of a regional dataset of intoxication records provide evidence that this is challenging, since occurring symptoms may not always match the textbook description due to regional distinctions, inter-rater variance, and institutional workflow. Computer-aided diagnosis (CADx) can provide decision support, but approaches so far do not consider additional information of the reported cases like age or gender, despite their potential value towards a correct diagnosis. In this work, we propose a new machine learning based CADx method which fuses symptoms and meta information of the patients using graph convolutional networks. We further propose a novel symptom matching method that allows the effective incorporation of prior knowledge into the learning process and evidently stabilizes the poison prediction. We validate our method against 10 medical doctors with different experience diagnosing intoxication cases for 10 different toxins from the PCC in Munich and show our method's superiority in performance for poison prediction.

Gerome Vivar, Kamilia Mullakaeva, Andreas Zwergal et al.

Computer-aided diagnosis (CADx) algorithms in medicine provide patient-specific decision support for physicians. These algorithms are usually applied after full acquisition of high-dimensional multimodal examination data, and often assume feature-completeness. This, however, is rarely the case due to examination costs, invasiveness, or a lack of indication. A sub-problem in CADx, which to our knowledge has received very little attention among the CADx community so far, is to guide the physician during the entire peri-diagnostic workflow, including the acquisition stage. We model the following question, asked from a physician's perspective: "Given the evidence collected so far, which examination should I perform next, in order to achieve the most accurate and efficient diagnostic prediction?". In this work, we propose a novel approach which is enticingly simple: use dropout at the input layer, and integrated gradients of the trained network at test-time to attribute feature importance dynamically. We validate and explain the effectiveness of our proposed approach using two public medical and two synthetic datasets. Results show that our proposed approach is more cost- and feature-efficient than prior approaches and achieves a higher overall accuracy. This directly translates to less unnecessary examinations for patients, and a quicker, less costly and more accurate decision support for the physician.

Luca Cosmo, Anees Kazi, Seyed-Ahmad Ahmadi et al.

Recently, Graph Convolutional Networks (GCNs) have proven to be a powerful machine learning tool for Computer-Aided Diagnosis (CADx) and disease prediction. A key component in these models is to build a population graph, where the graph adjacency matrix represents pair-wise patient similarities. Until now, the similarity metrics have been defined manually, usually based on meta-features like demographics or clinical scores. The definition of the metric, however, needs careful tuning, as GCNs are very sensitive to the graph structure. In this paper, we demonstrate for the first time in the CADx domain that it is possible to learn a single, optimal graph towards the GCN's downstream task of disease classification. To this end, we propose a novel, end-to-end trainable graph learning architecture for dynamic and localized graph pruning. Unlike commonly employed spectral GCN approaches, our GCN is spatial and inductive, and can thus infer previously unseen patients as well. We demonstrate significant classification improvements with our learned graph on two CADx problems in medicine. We further explain and visualize this result using an artificial dataset, underlining the importance of graph learning for more accurate and robust inference with GCNs in medical applications.

Anees Kazi, Luca Cosmo, Seyed-Ahmad Ahmadi et al.

Graph deep learning has recently emerged as a powerful ML concept allowing to generalize successful deep neural architectures to non-Euclidean structured data. Such methods have shown promising results on a broad spectrum of applications ranging from social science, biomedicine, and particle physics to computer vision, graphics, and chemistry. One of the limitations of the majority of current graph neural network architectures is that they are often restricted to the transductive setting and rely on the assumption that the underlying graph is {\em known} and {\em fixed}. Often, this assumption is not true since the graph may be noisy, or partially and even completely unknown. In such cases, it would be helpful to infer the graph directly from the data, especially in inductive settings where some nodes were not present in the graph at training time. Furthermore, learning a graph may become an end in itself, as the inferred structure may provide complementary insights next to the downstream task. In this paper, we introduce Differentiable Graph Module (DGM), a learnable function that predicts edge probabilities in the graph which are optimal for the downstream task. DGM can be combined with convolutional graph neural network layers and trained in an end-to-end fashion. We provide an extensive evaluation of applications from the domains of healthcare (disease prediction), brain imaging (age prediction), computer graphics (3D point cloud segmentation), and computer vision (zero-shot learning). We show that our model provides a significant improvement over baselines both in transductive and inductive settings and achieves state-of-the-art results.

Gerome Vivar, Hendrik Burwinkel, Anees Kazi et al.

Clinical diagnostic decision making and population-based studies often rely on multi-modal data which is noisy and incomplete. Recently, several works proposed geometric deep learning approaches to solve disease classification, by modeling patients as nodes in a graph, along with graph signal processing of multi-modal features. Many of these approaches are limited by assuming modality- and feature-completeness, and by transductive inference, which requires re-training of the entire model for each new test sample. In this work, we propose a novel inductive graph-based approach that can generalize to out-of-sample patients, despite missing features from entire modalities per patient. We propose multi-modal graph fusion which is trained end-to-end towards node-level classification. We demonstrate the fundamental working principle of this method on a simplified MNIST toy dataset. In experiments on medical data, our method outperforms single static graph approach in multi-modal disease classification.

Hendrik Burwinkel, Anees Kazi, Gerome Vivar et al.

Recently, Geometric Deep Learning (GDL) has been introduced as a novel and versatile framework for computer-aided disease classification. GDL uses patient meta-information such as age and gender to model patient cohort relations in a graph structure. Concepts from graph signal processing are leveraged to learn the optimal mapping of multi-modal features, e.g. from images to disease classes. Related studies so far have considered image features that are extracted in a pre-processing step. We hypothesize that such an approach prevents the network from optimizing feature representations towards achieving the best performance in the graph network. We propose a new network architecture that exploits an inductive end-to-end learning approach for disease classification, where filters from both the CNN and the graph are trained jointly. We validate this architecture against state-of-the-art inductive graph networks and demonstrate significantly improved classification scores on a modified MNIST toy dataset, as well as comparable classification results with higher stability on a chest X-ray image dataset. Additionally, we explain how the structural information of the graph affects both the image filters and the feature learning.

Anees Kazi, Shayan shekarforoush, S. Arvind krishna et al.

Geometric deep learning provides a principled and versatile manner for the integration of imaging and non-imaging modalities in the medical domain. Graph Convolutional Networks (GCNs) in particular have been explored on a wide variety of problems such as disease prediction, segmentation, and matrix completion by leveraging large, multimodal datasets. In this paper, we introduce a new spectral domain architecture for deep learning on graphs for disease prediction. The novelty lies in defining geometric 'inception modules' which are capable of capturing intra- and inter-graph structural heterogeneity during convolutions. We design filters with different kernel sizes to build our architecture. We show our disease prediction results on two publicly available datasets. Further, we provide insights on the behaviour of regular GCNs and our proposed model under varying input scenarios on simulated data.

Moustafa AboulAtta, Matthias Ossadnik, Seyed-Ahmad Ahmadi

Leveled Homomorphic Encryption (LHE) offers a potential solution that could allow sectors with sensitive data to utilize the cloud and securely deploy their models for remote inference with Deep Neural Networks (DNN). However, this application faces several obstacles due to the limitations of LHE. One of the main problems is the incompatibility of commonly used nonlinear functions in DNN with the operations supported by LHE, i.e. addition and multiplication. As common in LHE approaches, we train a model with a nonlinear function, and replace it with a low-degree polynomial approximation at inference time on private data. While this typically leads to approximation errors and loss in prediction accuracy, we propose a method that reduces this loss to small values or eliminates it entirely, depending on simple hyper-parameters. This is achieved by the introduction of a novel and elegantly simple Min-Max normalization scheme, which scales inputs to nonlinear functions into ranges with low approximation error. While being intuitive in its concept and trivial to implement, we empirically show that it offers a stable and effective approximation solution to nonlinear functions in DNN. In return, this can enable deeper networks with LHE, and facilitate the development of security- and privacy-aware analytics applications.

Gerome Vivar, Andreas Zwergal, Nassir Navab et al.

In large population-based studies and in clinical routine, tasks like disease diagnosis and progression prediction are inherently based on a rich set of multi-modal data, including imaging and other sensor data, clinical scores, phenotypes, labels and demographics. However, missing features, rater bias and inaccurate measurements are typical ailments of real-life medical datasets. Recently, it has been shown that deep learning with graph convolution neural networks (GCN) can outperform traditional machine learning in disease classification, but missing features remain an open problem. In this work, we follow up on the idea of modeling multi-modal disease classification as a matrix completion problem, with simultaneous classification and non-linear imputation of features. Compared to methods before, we arrange subjects in a graph-structure and solve classification through geometric matrix completion, which simulates a heat diffusion process that is learned and solved with a recurrent neural network. We demonstrate the potential of this method on the ADNI-based TADPOLE dataset and on the task of predicting the transition from MCI to Alzheimer's disease. With an AUC of 0.950 and classification accuracy of 87%, our approach outperforms standard linear and non-linear classifiers, as well as several state-of-the-art results in related literature, including a recently proposed GCN-based approach.

Atanas Mirchev, Seyed-Ahmad Ahmadi

In recent years, Generative Adversarial Networks (GAN) have emerged as a powerful method for learning the mapping from noisy latent spaces to realistic data samples in high-dimensional space. So far, the development and application of GANs have been predominantly focused on spatial data such as images. In this project, we aim at modeling of spatio-temporal sensor data instead, i.e. dynamic data over time. The main goal is to encode temporal data into a global and low-dimensional latent vector that captures the dynamics of the spatio-temporal signal. To this end, we incorporate auto-regressive RNNs, Wasserstein GAN loss, spectral norm weight constraints and a semi-supervised learning scheme into InfoGAN, a method for retrieval of meaningful latents in adversarial learning. To demonstrate the modeling capability of our method, we encode full-body skeletal human motion from a large dataset representing 60 classes of daily activities, recorded in a multi-Kinect setup. Initial results indicate competitive classification performance of the learned latent representations, compared to direct CNN/RNN inference. In future work, we plan to apply this method on a related problem in the medical domain, i.e. on recovery of meaningful latents in gait analysis of patients with vertigo and balance disorders.

Patrick Ferdinand Christ, Florian Ettlinger, Felix Grün et al.

Automatic segmentation of the liver and hepatic lesions is an important step towards deriving quantitative biomarkers for accurate clinical diagnosis and computer-aided decision support systems. This paper presents a method to automatically segment liver and lesions in CT and MRI abdomen images using cascaded fully convolutional neural networks (CFCNs) enabling the segmentation of a large-scale medical trial or quantitative image analysis. We train and cascade two FCNs for a combined segmentation of the liver and its lesions. In the first step, we train a FCN to segment the liver as ROI input for a second FCN. The second FCN solely segments lesions within the predicted liver ROIs of step 1. CFCN models were trained on an abdominal CT dataset comprising 100 hepatic tumor volumes. Validations on further datasets show that CFCN-based semantic liver and lesion segmentation achieves Dice scores over 94% for liver with computation times below 100s per volume. We further experimentally demonstrate the robustness of the proposed method on an 38 MRI liver tumor volumes and the public 3DIRCAD dataset.

Patrick Ferdinand Christ, Florian Ettlinger, Georgios Kaissis et al.

Automatic non-invasive assessment of hepatocellular carcinoma (HCC) malignancy has the potential to substantially enhance tumor treatment strategies for HCC patients. In this work we present a novel framework to automatically characterize the malignancy of HCC lesions from DWI images. We predict HCC malignancy in two steps: As a first step we automatically segment HCC tumor lesions using cascaded fully convolutional neural networks (CFCN). A 3D neural network (SurvivalNet) then predicts the HCC lesions' malignancy from the HCC tumor segmentation. We formulate this task as a classification problem with classes being "low risk" and "high risk" represented by longer or shorter survival times than the median survival. We evaluated our method on DWI of 31 HCC patients. Our proposed framework achieves an end-to-end accuracy of 65% with a Dice score for the automatic lesion segmentation of 69% and an accuracy of 68% for tumor malignancy classification based on expert annotations. We compared the SurvivalNet to classical handcrafted features such as Histogram and Haralick and show experimentally that SurvivalNet outperforms the handcrafted features in HCC malignancy classification. End-to-end assessment of tumor malignancy based on our proposed fully automatic framework corresponds to assessment based on expert annotations with high significance (p>0.95).

Patrick Ferdinand Christ, Mohamed Ezzeldin A. Elshaer, Florian Ettlinger et al.

Automatic segmentation of the liver and its lesion is an important step towards deriving quantitative biomarkers for accurate clinical diagnosis and computer-aided decision support systems. This paper presents a method to automatically segment liver and lesions in CT abdomen images using cascaded fully convolutional neural networks (CFCNs) and dense 3D conditional random fields (CRFs). We train and cascade two FCNs for a combined segmentation of the liver and its lesions. In the first step, we train a FCN to segment the liver as ROI input for a second FCN. The second FCN solely segments lesions from the predicted liver ROIs of step 1. We refine the segmentations of the CFCN using a dense 3D CRF that accounts for both spatial coherence and appearance. CFCN models were trained in a 2-fold cross-validation on the abdominal CT dataset 3DIRCAD comprising 15 hepatic tumor volumes. Our results show that CFCN-based semantic liver and lesion segmentation achieves Dice scores over 94% for liver with computation times below 100s per volume. We experimentally demonstrate the robustness of the proposed method as a decision support system with a high accuracy and speed for usage in daily clinical routine.

Fausto Milletari, Nassir Navab, Seyed-Ahmad Ahmadi

Convolutional Neural Networks (CNNs) have been recently employed to solve problems from both the computer vision and medical image analysis fields. Despite their popularity, most approaches are only able to process 2D images while most medical data used in clinical practice consists of 3D volumes. In this work we propose an approach to 3D image segmentation based on a volumetric, fully convolutional, neural network. Our CNN is trained end-to-end on MRI volumes depicting prostate, and learns to predict segmentation for the whole volume at once. We introduce a novel objective function, that we optimise during training, based on Dice coefficient. In this way we can deal with situations where there is a strong imbalance between the number of foreground and background voxels. To cope with the limited number of annotated volumes available for training, we augment the data applying random non-linear transformations and histogram matching. We show in our experimental evaluation that our approach achieves good performances on challenging test data while requiring only a fraction of the processing time needed by other previous methods.

Fausto Milletari, Seyed-Ahmad Ahmadi, Christine Kroll et al.

In this work we propose a novel approach to perform segmentation by leveraging the abstraction capabilities of convolutional neural networks (CNNs). Our method is based on Hough voting, a strategy that allows for fully automatic localisation and segmentation of the anatomies of interest. This approach does not only use the CNN classification outcomes, but it also implements voting by exploiting the features produced by the deepest portion of the network. We show that this learning-based segmentation method is robust, multi-region, flexible and can be easily adapted to different modalities. In the attempt to show the capabilities and the behaviour of CNNs when they are applied to medical image analysis, we perform a systematic study of the performances of six different network architectures, conceived according to state-of-the-art criteria, in various situations. We evaluate the impact of both different amount of training data and different data dimensionality (2D, 2.5D and 3D) on the final results. We show results on both MRI and transcranial US volumes depicting respectively 26 regions of the basal ganglia and the midbrain.