holds the Chair of Computer Aided Medical Procedures & Augmented Reality at TU Munich.
His research focuses on computer-aided medical procedures and augmented reality. The work involves developing technologies to improve the quality of medical intervention and bridges the gap between medicine and computer science.
Ultrasound is widely used in medical diagnostics allowing for accessible and powerful imaging but suffers from resolution limitations due to diffraction and the finite aperture of the imaging system, which restricts diagnostic use. The impulse function of an ultrasound imaging system is called the point spread function (PSF), which is convolved with the spatial distribution of reflectors in the image formation process. Recovering high-resolution reflector distributions by removing image distortions induced by the convolution process improves image clarity and detail. Conventionally, deconvolution techniques attempt to rectify the imaging system’s dependent PSF, working directly on the radio-frequency (RF) data. However, RF data is often not readily accessible. Therefore, we introduce a physics-based deconvolution process using a modeled PSF, working directly on the more commonly available B-mode images. By leveraging Implicit Neural Representations (INRs), we learn a continuous mapping from spatial locations to their respective echogenicity values, effectively compensating for the discretized image space. Our contribution consists of a novel methodology for retrieving a continuous echogenicity map directly from a B-mode image through a differentiable physics-based rendering pipeline for ultrasound resolution enhancement. We qualitatively and quantitatively evaluate our approach on synthetic data, demonstrating improvements over traditional methods in metrics such as PSNR and SSIM. Furthermore, we show qualitative enhancements on an ultrasound phantom and an in-vivo acquisition of a carotid artery.
In radiation therapy (RT), an accurate delineation of the regions of interest (ROI) and organs at risk (OAR) allows for a more targeted irradiation with reduced side effects. The current clinical workflow for combined MR-linear accelerator devices (MR-linacs) requires the acquisition of a planning MR volume (MR-P), in which the ROI and OAR are accurately segmented by the clinical team. These segmentation maps (S-P) are transferred to the MR acquired on the day of the RT fraction (MR-Fx) using registration, followed by time-consuming manual corrections. The goal of this paper is to enable accurate automatic segmentation of MR-Fx using S-P without clinical workflow disruption. We propose a novel UNet-based architecture, CloverNet, that takes as inputs MR-Fx and S-P in two separate encoder branches, whose latent spaces are concatenated in the bottleneck to generate an improved segmentation of MP-Fx. CloverNet improves the absolute Dice Score by 3.73% (relative +4.34%, p<0.001) when compared with conventional 3D UNet. Moreover, we believe this approach is potentially applicable to other longitudinal use cases in which a prior segmentation of the ROI is available.
Surgical data science (SDS) is a field that analyzes patient data before, during, and after surgery to improve surgical outcomes and skills. However, surgical data is scarce, heterogeneous, and complex, which limits the applicability of existing machine learning methods. In this work, we introduce the novel task of future video generation in laparoscopic surgery. This task can augment and enrich the existing surgical data and enable various applications, such as simulation, analysis, and robot-aided surgery. Ultimately, it involves not only understanding the current state of the operation but also accurately predicting the dynamic and often unpredictable nature of surgical procedures. Our proposed method, VISAGE (VIdeo Synthesis using Action Graphs for Surgery), leverages the power of action scene graphs to capture the sequential nature of laparoscopic procedures and utilizes diffusion models to synthesize temporally coherent video sequences. VISAGE predicts the future frames given only a single initial frame, and the action graph triplets. By incorporating domain-specific knowledge through the action graph, VISAGE ensures the generated videos adhere to the expected visual and motion patterns observed in real laparoscopic procedures. The results of our experiments demonstrate high-fidelity video generation for laparoscopy procedures, which enables various applications in SDS.
Graph-based holistic scene representations facilitate surgical workflow understanding and have recently demonstrated significant success. However, this task is often hindered by the limited availability of densely annotated surgical scene data. In this work, we introduce an end-to-end framework for the generation and optimization of surgical scene graphs on a downstream task. Our approach leverages the flexibility of graph-based spectral clustering and the generalization capability of foundation models to generate unsupervised scene graphs with learnable properties. We reinforce the initial spatial graph with sparse temporal connections using local matches between consecutive frames to predict temporally consistent clusters across a temporal neighborhood. By jointly optimizing the spatiotemporal relations and node features of the dynamic scene graph with the downstream task of phase segmentation, we address the costly and annotation-burdensome task of semantic scene comprehension and scene graph generation in surgical videos using only weak surgical phase labels. Further, by incorporating effective intermediate scene representation disentanglement steps within the pipeline, our solution outperforms the SOTA on the CATARACTS dataset by 8% accuracy and 10% F1 score in surgical workflow recognition.
Deep learning (DL) methods typically require large datasets to effectively learn data distributions. However, in the medical field, data is often limited in quantity, and acquiring labeled data can be costly. To mitigate this data scarcity, data augmentation techniques are commonly employed. Among these techniques, generative models play a pivotal role in expanding datasets. However, when it comes to ultrasound (US) imaging, the authenticity of generated data often diminishes due to the oversight of ultrasound physics.
We propose a novel approach to improve the quality of generated US images by introducing a physics-based diffusion model that is specifically designed for this image modality. The proposed model incorporates an US-specific scheduler scheme that mimics the natural behavior of sound wave propagation in ultrasound imaging. Our analysis demonstrates how the proposed method aids in modeling the attenuation dynamics in US imaging. We present both qualitative and quantitative results based on standard generative model metrics, showing that our proposed method results in overall more plausible images.
Ultrasound imaging is challenging to interpret due to non-uniform intensities, low contrast, and inherent artifacts, necessitating extensive training for non-specialists. Advanced representation with clear tissue structure separation could greatly assist clinicians in mapping underlying anatomy and distinguishing between tissue layers. Decomposing an image into semantically meaningful segments is mainly achieved using supervised segmentation algorithms. Unsupervised methods are beneficial, as acquiring large labeled datasets is difficult and costly, but despite their advantages, they still need to be explored in ultrasound. This paper proposes a novel unsupervised deep learning strategy tailored to ultrasound to obtain easily interpretable tissue separations. We integrate key concepts from unsupervised deep spectral methods, which combine spectral graph theory with deep learning methods. We utilize self-supervised transformer features for spectral clustering to generate meaningful segments based on ultrasound-specific metrics and shape and positional priors, ensuring semantic consistency across the dataset. We evaluate our unsupervised deep learning strategy on three ultrasound datasets, showcasing qualitative results across anatomical contexts without label requirements. We also conduct a comparative analysis against other clustering algorithms to demonstrate superior segmentation performance, boundary preservation, and label consistency.
Nuclei semantic segmentation is a key component for advancing machine learning and deep learning applications in digital pathology. However, most existing segmentation models are trained and tested on high-quality data acquired with expensive equipment, such as whole slide scanners, which are not accessible to most pathologists in developing countries. These pathologists rely on low-resource data acquired with low-precision microscopes, smartphones, or digital cameras, which have different characteristics and challenges than high-resource data. Therefore, there is a gap between the state-of-the-art segmentation models and the real-world needs of low-resource settings. This work aims to bridge this gap by presenting the first fully annotated African multi-organ dataset for histopathology nuclei semantic segmentation acquired with a low-precision microscope. We also evaluate state-of-the-art segmentation models, including spectral feature extraction encoder and vision transformer-based models, and stain normalization techniques for color normalization of Hematoxylin and Eosin-stained histopathology slides. Our results provide important insights for future research on nuclei histopathology segmentation with low-resource data.
Every day, countless surgeries are performed worldwide, each within the distinct settings of operating rooms (ORs) that vary not only in their setups but also in the personnel, tools, and equipment used. This inherent diversity poses a substantial challenge for achieving a holistic understanding of the OR, as it requires models to generalize beyond their initial training datasets. To reduce this gap, we introduce ORacle, an advanced vision-language model designed for holistic OR domain modeling, which incorporates multi-view and temporal capabilities and can leverage external knowledge during inference, enabling it to adapt to previously unseen surgical scenarios. This capability is further enhanced by our novel data augmentation framework, which significantly diversifies the training dataset, ensuring ORacle’s proficiency in applying the provided knowledge effectively. In rigorous testing, in scene graph generation, and downstream tasks on the 4D-OR dataset, ORacle not only demonstrates state-of-the-art performance but does so requiring less data than existing models. Furthermore, its adaptability is displayed through its ability to interpret unseen views, actions, and appearances of tools and equipment. This demonstrates ORacle’s potential to significantly enhance the scalability and affordability of OR domain modeling and opens a pathway for future advancements in surgical data science.
In emergency departments, rural hospitals, or clinics in less developed regions, clinicians often lack fast image analysis by trained radiologists, which can have a detrimental effect on patients’ healthcare. Large Language Models (LLMs) have the potential to alleviate some pressure from these clinicians by providing insights that can help them in their decision-making. While these LLMs achieve high test results on medical exams showcasing their great theoretical medical knowledge, they tend not to follow medical guidelines. In this work, we introduce a new approach for zero-shot guideline-driven decision support. We model a system of multiple LLM agents augmented with a contrastive vision-language model that collaborate to reach a patient diagnosis. After providing the agents with simple diagnostic guidelines, they will synthesize prompts and screen the image for findings following these guidelines. Finally, they provide understandable chain-of-thought reasoning for their diagnosis, which is then self-refined to consider inter-dependencies between diseases. As our method is zero-shot, it is adaptable to settings with rare diseases, where training data is limited, but expert-crafted disease descriptions are available. We evaluate our method on two chest X-ray datasets, CheXpert and ChestX-ray 14 Longtail, showcasing performance improvement over existing zero-shot methods and generalizability to rare diseases.
We present a new model for deformable image registration, which learns in an unsupervised way a data-specific similarity metric. The proposed method consists of two neural networks, one that maps pairs of input images to transformations which align them, and one that provides the similarity metric whose maximisation guides the image alignment. We parametrise the similarity metric as an energy-based model, which is simple to train and allows us to improve the accuracy of image registration compared to other models with learnt similarity metrics by taking advantage of a more general mathematical formulation, as well as larger datasets. We also achieve substantial improvement in the accuracy of inter-patient image registration on MRI scans from the OASIS dataset compared to models that rely on traditional functions.
Recent advances in generative models for medical imaging have shown promise in representing multiple modalities. However, the variability in modality availability across datasets limits the general applicability of the synthetic data they produce. To address this, we present a novel physics-informed generative model capable of synthesizing a variable number of brain MRI modalities, including those not present in the original dataset. Our approach utilizes latent diffusion models and a two-step generative process: first, unobserved physical tissue property maps are synthesized using a latent diffusion model, and then these maps are combined with a physical signal model to generate the final MRI scan. Our experiments demonstrate the efficacy of this approach in generating unseen MR contrasts and preserving physical plausibility. Furthermore, we validate the distributions of generated tissue properties by comparing them to those measured in real brain tissue.
We present EchoScene, an interactive and controllable generative model that generates 3D indoor scenes on scene graphs. EchoScene leverages a dual-branch diffusion model that dynamically adapts to scene graphs. Existing methods struggle to handle scene graphs due to varying numbers of nodes, multiple edge combinations, and manipulator-induced node-edge operations. EchoScene overcomes this by associating each node with a denoising process and enables collaborative information exchange, enhancing controllable and consistent generation aware of global constraints. This is achieved through an information echo scheme in both shape and layout branches. At every denoising step, all processes share their denoising data with an information exchange unit that combines these updates using graph convolution. The scheme ensures that the denoising processes are influenced by a holistic understanding of the scene graph, facilitating the generation of globally coherent scenes. The resulting scenes can be manipulated during inference by editing the input scene graph and sampling the noise in the diffusion model. Extensive experiments validate our approach, which maintains scene controllability and surpasses previous methods in generation fidelity. Moreover, the generated scenes are of high quality and thus directly compatible with off-the-shelf texture generation. Our code and models are open-sourced.
In minimally invasive endovascular procedures, contrast-enhanced angiography remains the most robust imaging technique. However, it is at the expense of the patient and clinician’s health due to prolonged radiation exposure. As an alternative, interventional ultrasound has notable benefits such as being radiation-free, fast to deploy, and having a small footprint in the operating room. Yet, ultrasound is hard to interpret, and highly prone to artifacts and noise. Additionally, interventional radiologists must undergo extensive training before they become qualified to diagnose and treat patients effectively, leading to a shortage of staff, and a lack of open-source datasets. In this work, we seek to address both problems by introducing a self-supervised deep learning architecture to segment catheters in longitudinal ultrasound images, without demanding any labeled data. The network architecture builds upon AiAReSeg, a segmentation transformer built with the Attention in Attention mechanism, and is capable of learning feature changes across time and space. To facilitate training, we used synthetic ultrasound data based on physics-driven catheter insertion simulations, and translated the data into a unique CT-Ultrasound common domain, CACTUSS, to improve the segmentation performance. We generated ground truth segmentation masks by computing the optical flow between adjacent frames using FlowNet2, and performed thresholding to obtain a binary map estimate. Finally, we validated our model on a test dataset, consisting of unseen synthetic data and images collected from silicon aorta phantoms, thus demonstrating its potential for applications to clinical data in the future.
Category-level object pose estimation, aiming to predict the 6D pose and 3D size of objects from known categories, typically struggles with large intra-class shape variation. Existing works utilizing mean shapes often fall short of cap-turing this variation. To address this issue, we present Sec-ondPose, a novel approach integrating object-specific ge-ometric features with semantic category priors from DI-NOv2. Leveraging the advantage of DINOv2 in providing SE(3)-consistent semantic features, we hierarchically extract two types of SE(3)-invariant geometric features to further encapsulate local-to-global object-specific information. These geometric features are then point-aligned with DINOv2 features to establish a consistent object represen-tation under SE(3) transformations, facilitating the map-ping from camera space to the pre-defined canonical space, thus further enhancing pose estimation. Extensive exper-iments on NOCS-REAL275 demonstrate that SecondPose achieves a 12.4% leap forward over the state-of-the-art. Moreover, on a more complex dataset HouseCat6D which provides photometrically challenging objects, SecondPose still surpasses other competitors by a large margin.
Recent learning methods for object pose estimation require resource-intensive training for each individual object instance or category, hampering their scalability in real applications when confronted with previously unseen objects. In this paper, we propose MatchU, a Fuse-Describe-Match strategy for 6D pose estimation from RGB-D images. MatchU is a generic approach that fuses 2D texture and 3D geometric cues for 6D pose prediction of unseen objects. We rely on learning geometric 3D descriptors that are rotation-invariant by design. By encoding pose-agnostic geometry, the learned descriptors naturally generalize to unseen objects and capture symmetries. To tackle ambiguous associations using 3D geometry only, we fuse additional RGB information into our descriptor. This is achieved through a novel attention-based mechanism that fuses cross-modal information, together with a matching loss that leverages the latent space learned from RGB data to guide the descriptor learning process. Extensive experiments reveal the generalizability of both the RGB-D fusion strategy as well as the descriptor efficacy. Benefiting from the novel designs, MatchU surpasses all existing methods by a significant margin in terms of both accuracy and speed, even without the requirement of expensive re-training or rendering.
Estimating 6D object poses is a major challenge in 3D computer vision. Building on successful instance-level approaches, research is shifting towards category-level pose estimation for practical applications. Current category-level datasets, however, fall short in annotation quality and pose variety. Addressing this, we introduce HouseCat6D, a new category-level 6D pose dataset. It features 1) multi-modality with Polarimetric RGB and Depth (RGBD+P), 2) encompasses 194 diverse objects across 10 household cat-egories, including two photometrically challenging ones, and 3) provides high-quality pose annotations with an error range of only 1.35 mm to 1.74 mm. The dataset also includes 4) 41 large-scale scenes with comprehensive view-point and occlusion coverage,5) a checkerboard-free en-vironment, and 6) dense 6D parallel-jaw robotic grasp annotations. Additionally, we present benchmark results for leading category-level pose estimation networks.
Ultrasound (US) imaging is widely used in diagnosing and staging abdominal diseases due to its lack of non-ionizing radiation and prevalent availability. However, significant inter-operator variability and inconsistent image acquisition hinder the widespread adoption of extensive screening programs. Robotic ultrasound systems have emerged as a promising solution, offering standardized acquisition protocols and the possibility of automated acquisition. Additionally, these systems enable access to 3D data via robotic tracking, enhancing volumetric reconstruction for improved ultrasound interpretation and precise disease diagnosis.However, the interpretability of 3D US reconstruction of abdominal images can be affected by the patient’s breathing motion. This study introduces a method to compensate for breathing motion in 3D US compounding by leveraging implicit neural representations. Our approach employs a robotic ultrasound system for automated screenings. To demonstrate the method’s effectiveness, we evaluate our proposed method for the diagnosis and monitoring of abdominal aorta aneurysms as a representative use case.Our experiments demonstrate that our proposed pipeline facilitates robust automated robotic acquisition, mitigating artifacts from breathing motion, and yields smoother 3D reconstructions for enhanced screening and medical diagnosis.
Segmenting ultrasound images is important for precise area and/or volume calculations, ensuring reliable diagnosis and effective treatment evaluation for diseases. Recently, many segmentation methods have been proposed and shown impressive performance. However, currently, there is no deeper understanding of how networks segment target regions or how they define the boundaries. In this paper, we present a new approach that analyzes ultrasound segmentation networks in terms of learned borders because border delimitation is challenging in ultrasound.
Ultrasound (US) imaging, while advantageous for its radiation-free nature, is challenging to interpret due to only partially visible organs and a lack of complete 3D information. While performing US-based diagnosis or investigation, medical professionals therefore create a mental map of the 3D anatomy. In this work, we aim to replicate this process and enhance the visual representation of anatomical structures.
Learning to learn is a powerful paradigm that enables machine learning models to leverage the previously learned features for new tasks and domains more effectively. This thesis explores different aspects of learning to learn from data, models, and semantics, and shows how they can enhance various computer vision and medical imaging tasks. In the first part of the thesis, we present novel and fundamental research on learning to learn from data, and in the second part, we investigate the use of high-level semantics in generative models.
Controllable scene synthesis aims to create interactive environments for numerous industrial use cases. Scene graphs provide a highly suitable interface to facilitate these applications by abstracting the scene context in a compact manner. Existing methods, reliant on retrieval from extensive databases or pre-trained shape embeddings, often overlook scene-object and object-object relationships, leading to inconsistent results due to their limited generation capacity. To address this issue, we present CommonScenes, a fully generative model that converts scene graphs into corresponding controllable 3D scenes, which are semantically realistic and conform to commonsense. Our pipeline consists of two branches, one predicting the overall scene layout via a variational auto-encoder and the other generating compatible shapes via latent diffusion, capturing global scene-object and local inter-object relationships in the scene graph while preserving shape diversity. The generated scenes can be manipulated by editing the input scene graph and sampling the noise in the diffusion model. Due to the lack of a scene graph dataset offering high-quality object-level meshes with relations, we also construct SG-FRONT, enriching the off-the-shelf indoor dataset 3D-FRONT with additional scene graph labels. Extensive experiments are conducted on SG-FRONT, where CommonScenes shows clear advantages over other methods regarding generation consistency, quality, and diversity. Codes and the dataset are available on the website.
Aligning 2D ultrasound images with 3D CT scans of the liver holds significant clinical value in enhancing diagnostic precision, surgical planning, and treatment delivery. Conventional approaches primarily rely on optimization techniques, which often have a limited capture range and are susceptible to initialization errors. To address these limitations, we define the problem as “probe pose regression” and leverage deep learning for a more robust and efficient solution for liver US-CT registration without access to paired data. The proposed method is a three-part framework that combines ultrasound rendering, generative model and pose regression. In the first stage, we exploit a differentiable ultrasound rendering model designed to synthesize ultrasound images given segmentation labels. We let the downstream task optimize the rendering parameters, enhancing the performance of the overall method. In the second stage, a generative model bridges the gap between real and rendered ultrasound images, enabling application on real B-mode images. Finally, we use a patient-specific pose regression network, trained self-supervised with only synthetic images and their known poses. We use ultrasound, and CT scans from a dual-modality human abdomen phantom to validate the proposed method.
Our experimental results indicate that the proposed method can estimate probe poses within an acceptable error margin, which can later be fine-tuned using conventional methods. This capability confirms that the proposed framework can serve as a reliable initialization step for US-CT fusion and achieve fully automated US-CT fusion when coupled with conventional methods.
Inpainting has recently been employed as a successful deep-learning technique for unsupervised model discovery in medical image analysis by taking advantage of the strong priors learned by models to reconstruct the structure and texture of missing parts in images. Even though the learned features depend on the masks as well as the images, the masks used for inpainting are typically random and independent of the dataset, due to the unpredictability of the content of images, i.e., different objects and shapes can appear in different locations in images. However, this is rarely the case for medical imaging data since they are obtained from similar anatomies. Still, random square masks are the most popular technique for inpainting in medical imaging. In this work, we propose a pipeline to generate, position and sample the masks to efficiently learn the shape and structures of the anatomy and generate a myriad of diverse anatomy-aware masks, aiding the model in learning the statistical shape prior to the topology of the organs of interest. We demonstrate the impact of our approach compared to other masking mechanisms in the reconstruction of anatomy. We compare the effectiveness of our proposed masking approach over square-shaped masks, which are traditionally used in medical imaging, and irregular shape masks, which are used in SOTA inpainting literature.
Text-conditioned image generation has made significant progress in recent years with generative adversarial networks and more recently, diffusion models. While diffusion models conditioned on text prompts have produced impressive and high-quality images, accurately representing complex text prompts such as the number of instances of a specific object remains challenging.To address this limitation, we propose a novel guidance approach for the sampling process in the diffusion model that leverages bounding box and segmentation map information at inference time without additional training data. Through a novel loss in the sampling process, our approach guides the model with semantic features from CLIP embeddings and enforces geometric constraints, leading to high-resolution images that accurately represent the scene. To obtain bounding box and segmentation map information, we structure the text prompt as a scene graph and enrich the nodes with CLIP embeddings. Our proposed model achieves state-of-the-art performance on two public benchmarks for image generation from scene graphs, surpassing both scene graph to image and text-based diffusion models in various metrics. Our results demonstrate the effectiveness of incorporating bounding box and segmentation map guidance in the diffusion model sampling process for more accurate text-to-image generation.
Although purely transformer-based architectures pretrained on large datasets are introduced as foundation models for general computer vision tasks, hybrid models that incorporate combinations of convolution and transformer blocks showed state-of-the-art performance in more specialized tasks. Nevertheless, despite the performance gain of both pure and hybrid transformer-based architectures compared to convolutional networks, their high training cost and complexity make it challenging to use them in real scenarios. In this work, we propose a novel and simple architecture based on only convolutional layers and show that by just taking advantage of the attention map visualizations obtained from a self-supervised pretrained vision transformer network, complex transformer-based networks, and even 3D architectures are outperformed with much fewer 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 in the other branch. The results of our experiments on medical imaging datasets show that the extracted attention map visualizations from the attention heads of a pre-trained transformer architecture combined with the image provide strong prior knowledge for a pure CNN architecture to outperform CNN-based and transformer-based architectures.
Recent advances in semantic scene understanding have underscored its growing significance in the field of computer vision. Enhanced representations can be achieved by incorporating semantic information derived from textual data and applying it to generative models for scene modeling. Nevertheless, the features extracted from text prompts may not seamlessly model a scene.
Scene graphs offer a robust solution to address this challenge, serving as a powerful representation for semantic image generation and manipulation. In this study, we delve into the utilization of scene graphs for this purpose and propose novel methodologies to augment both the representation and learning processes involved in image generation and manipulation.
For image generation, we examine meta-learning for producing images in unprecedented scenes and refine the generated images using an autoregressive scene graph generation model. In terms of image manipulation, we put forth a novel self-supervised method that eliminates the need for paired before-and-after data. Additionally, we boost image manipulation performance by disentangling latent and graph representations in a self-supervised manner.
By evaluating the efficacy of our proposed approaches on a diverse range of publicly available benchmarks, we demonstrate their superiority, ultimately achieving state-of-the-art performance in the domain of semantic image generation and manipulation.
Although the preservation of shape continuity and physiological anatomy is a natural assumption in the segmentation of medical images, it is often neglected by deep learning methods that mostly aim for the statistical modeling of input data as pixels rather than interconnected structures. In biological structures, however, organs are not separate entities; for example, in reality, a severed vessel is an indication of an underlying problem, but traditional segmentation models are not designed to strictly enforce the continuity of anatomy, potentially leading to inaccurate medical diagnoses. To address this issue, we propose a graph-based approach that enforces the continuity and connectivity of anatomical topology in medical images. Our method encodes the continuity of shapes as a graph constraint, ensuring that the network’s predictions maintain this continuity. We evaluate our method on two public benchmarks on retinal vessel segmentation, showing significant improvements in connectivity metrics compared to traditional methods while getting better or on-par performance on segmentation metrics.
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.
Feature attribution is arguably the predominant approach for illuminating black-box neural networks. This dissertation rethinks feature attribution by leveraging critical neural pathways, identifying input features with predictive information, and evaluating feature attribution using the neural network model. The dissertation also rethinks feature attribution for the explanation of medical imaging models.
Graph representation of objects and their relations in a scene, known as a scene graph, provides a precise and discernible interface to manipulate a scene by modifying the nodes or the edges in the graph. Although existing works have shown promising results in modifying the placement and pose of objects, scene manipulation often leads to losing some visual characteristics like the appearance or identity of objects. In this work, we propose DisPositioNet, a model that learns a disentangled representation for each object for the task of image manipulation using scene graphs in a self-supervised manner. Our framework enables the disentanglement of the variational latent embeddings as well as the feature representation in the graph. In addition to producing more realistic images due to the decomposition of features like pose and identity, our method takes advantage of the probabilistic sampling in the intermediate features to generate more diverse images in object replacement or addition tasks. The results of our experiments show that disentangling the feature representations in the latent manifold of the model outperforms the previous works qualitatively and quantitatively on two public benchmarks.
The lack of sufficient annotated image data is a common issue in medical image segmentation. For some organs and densities, the annotation may be scarce, leading to poor model training convergence, while other organs have plenty of annotated data. In this work, we present MetaMedSeg, a gradient-based meta-learning algorithm that redefines the meta-learning task for the volumetric medical data with the goal of capturing the variety between the slices. We also explore different weighting schemes for gradients aggregation, arguing that different tasks might have different complexity and hence, contribute differently to the initialization. We propose an importance-aware weighting scheme to train our model. In the experiments, we evaluate our method on the medical decathlon dataset by extracting 2D slices from CT and MRI volumes of different organs and performing semantic segmentation. The results show that our proposed volumetric task definition leads to up to improvement in terms of IoU compared to related baselines. The proposed update rule is also shown to improve the performance for complex scenarios where the data distribution of the target organ is very different from the source organs.
Federated learning (FL) is a distributed learning method that offers medical institutes the prospect of collaboration in a global model while preserving the privacy of their patients. Although most medical centers conduct similar medical imaging tasks, their differences, such as specializations, number of patients, and devices, lead to distinctive data distributions. Data heterogeneity poses a challenge for FL and the personalization of the local models. In this work, we investigate an adaptive hierarchical clustering method for FL to produce intermediate semi-global models, so clients with similar data distribution have the chance of forming a more specialized model. Our method forms several clusters consisting of clients with the most similar data distributions; then, each cluster continues to train separately. Inside the cluster, we use meta-learning to improve the personalization of the participants’ models. We compare the clustering approach with classical FedAvg and centralized training by evaluating our proposed methods on the HAM10k dataset for skin lesion classification with extreme heterogeneous data distribution. Our experiments demonstrate significant performance gain in heterogeneous distribution compared to standard FL methods in classification accuracy. Moreover, we show that the models converge faster if applied in clusters and outperform centralized training while using only a small subset of data.
Automated segmentation of retinal optical coherence tomography (OCT) images has become an important recent direction in machine learning for medical applications. We hypothesize that the anatomic structure of layers and their high-frequency variation in OCT images make retinal OCT a fitting choice for extracting spectral domain features and combining them with spatial domain features. In this work, we present Y-Net, an architecture that combines the frequency domain features with the image domain to improve the segmentation performance of OCT images. The results of this work demonstrate that the introduction of two branches, one for spectral and one for spatial domain features, brings very significant improvement in fluid segmentation performance and allows outperformance as compared to the well-known U-Net model. Our improvement was 13% on the fluid segmentation dice score and 1.9% on the average dice score. Finally, removing selected frequency ranges in the spectral domain demonstrates the impact of these features on the fluid segmentation outperformance.
Do black-box neural network models learn clinically relevant features for fracture diagnosis? The answer not only establishes reliability, quenches scientific curiosity, but also leads to explainable and verbose findings that can assist the radiologists in the final and increase trust. This work identifies the concepts networks use for vertebral fracture diagnosis in CT images. This is achieved by associating concepts to neurons highly correlated with a specific diagnosis in the dataset. The concepts are either associated with neurons by radiologists pre-hoc or are visualized during a specific prediction and left for the user’s interpretation. We evaluate which concepts lead to correct diagnosis and which concepts lead to false positives. The proposed frameworks and analysis pave the way for reliable and explainable vertebral fracture diagnosis.
One challenging property lurking in medical datasets is the imbalanced data distribution, where the frequency of the samples between the different classes is not balanced. Training a model on an imbalanced dataset can introduce unique challenges to the learning problem where a model is biased towards the highly frequent class. Many methods are proposed to tackle the distributional differences and the imbalanced problem. However, the impact of these approaches on the learned features is not well studied. In this paper, we look deeper into the internal units of neural networks to observe how handling data imbalance affects the learned features. We study several popular cost-sensitive approaches for handling data imbalance and analyze the feature maps of the convolutional neural networks from multiple perspectives: analyzing the alignment of salient features with pathologies and analyzing the pathology-related concepts encoded by the networks. Our study reveals differences and insights regarding the trained models that are not reflected by quantitative metrics such as AUROC and AP and show up only by looking at the models through a lens.
Inpainting has recently been proposed as a successful deep learning technique for unsupervised medical image model discovery. The masks used for inpainting are generally independent of the dataset and are not tailored to perform on different given classes of anatomy. In this work, we introduce a method for generating shape-aware masks for inpainting, which aims at learning the statistical shape prior. We hypothesize that although the variation of masks improves the generalizability of inpainting models, the shape of the masks should follow the topology of the organs of interest. Hence, we propose an unsupervised guided masking approach based on an off-the-shelf inpainting model and a superpixel over-segmentation algorithm to generate a wide range of shape-dependent masks. Experimental results on abdominal MR image reconstruction show the superiority of our proposed masking method over standard methods using square-shaped or dataset of irregular shape masks.
It is a mystery which input features contribute to a neural network’s output. Various explanation (feature attribution) methods are proposed in the literature to shed light on the problem. One peculiar observation is that these explanations (attributions) point to different features as being important. The phenomenon raises the question, which explanation to trust? We propose a framework for evaluating the explanations using the neural network model itself. The framework leverages the network to generate input features that impose a particular behavior on the output. Using the generated features, we devise controlled experimental setups to evaluate whether an explanation method conforms to an axiom. Thus we propose an empirical framework for axiomatic evaluation of explanation methods. We evaluate well-known and promising explanation solutions using the proposed framework. The framework provides a toolset to reveal properties and drawbacks within existing and future explanation solutions
The automation of chest X-ray reporting has garnered significant interest due to the time-consuming nature of the task. However, the clinical accuracy of free-text reports has proven challenging to quantify using natural language processing metrics, given the complexity of medical information, the variety of writing styles, and the potential for typos and inconsistencies. Structured reporting and standardized reports, on the other hand, can provide consistency and formalize the evaluation of clinical correctness. However, high-quality annotations for structured reporting are scarce. Therefore, we propose a method to predict clinical findings defined by sentences in structured reporting templates, which can be used to fill such templates. The approach involves training a contrastive language-image model using chest X-rays and related free-text radiological reports, then creating textual prompts for each structured finding and optimizing a classifier to predict clinical findings in the medical image. Results show that even with limited image-level annotations for training, the method can accomplish the structured reporting tasks of severity assessment of cardiomegaly and localizing pathologies in chest X-rays.
Freehand ultrasound imaging is an important medical imaging modality due to its ease of applicability and wide application spectrum. Still, modern ultrasound imaging is a largely passive imaging modality, and does not dynamically adapt to the physics in the medium of interest. This dissertation presents the application of physics-informed deep learning for ultrasound imaging applied to sound speed estimation.
One principal approach for illuminating a black-box neural network is feature attribution, i.e. identifying the importance of input features for the network’s prediction. The predictive information of features is recently proposed as a proxy for the measure of their importance. So far, the predictive information is only identified for latent features by placing an information bottleneck within the network. We propose a method to identify features with predictive information in the input domain. The method results in fine-grained identification of input features’ information and is agnostic to network architecture. The core idea of our method is leveraging a bottleneck on the input that only lets input features associated with predictive latent features pass through. We compare our method with several feature attribution methods using mainstream feature attribution evaluation experiments. The code is publicly available.
Generation of images from scene graphs is a promising direction towards explicit scene generation and manipulation. However, the images generated from the scene graphs lack quality, which in part comes due to high difficulty and diversity in the data. We propose MIGS (Meta Image Generation from Scene Graphs), a meta-learning based approach for few-shot image generation from graphs that enables adapting the model to different scenes and increases the image quality by training on diverse sets of tasks. By sampling the data in a task-driven fashion, we train the generator using meta-learning on different sets of tasks that are categorized based on the scene attributes. Our results show that using this meta-learning approach for the generation of images from scene graphs achieves state-of-the-art performance in terms of image quality and capturing the semantic relationships in the scene.
Despite recent advancements in single-domain or single-object image generation, it is still challenging to generate complex scenes containing diverse, multiple objects and their interactions. Scene graphs, composed of nodes as objects and directed-edges as relationships among objects, offer an alternative representation of a scene that is more semantically grounded than images. We hypothesize that a generative model for scene graphs might be able to learn the underlying semantic structure of real-world scenes more effectively than images, and hence, generate realistic novel scenes in the form of scene graphs. In this work, we explore a new task for the unconditional generation of semantic scene graphs. We develop a deep auto-regressive model called SceneGraphGen which can directly learn the probability distribution over labelled and directed graphs using a hierarchical recurrent architecture. The model takes a seed object as input and generates a scene graph in a sequence of steps, each step generating an object node, followed by a sequence of relationship edges connecting to the previous nodes. We show that the scene graphs generated by SceneGraphGen are diverse and follow the semantic patterns of real-world scenes. Additionally, we demonstrate the application of the generated graphs in image synthesis, anomaly detection and scene graph completion.
Convolutional neural networks are showing promise in the automatic diagnosis of thoracic pathologies on chest x-rays. Their black-box nature has sparked many recent works to explain the prediction via input feature attribution methods (aka saliency methods). However, input feature attribution methods merely identify the importance of input regions for the prediction and lack semantic interpretation of model behavior. In this work, we first identify the semantics associated with internal units (feature maps) of the network. We proceed to investigate the following questions; Does a regression model that is only trained with COVID-19 severity scores implicitly learn visual patterns associated with thoracic pathologies? Does a network that is trained on weakly labeled data (e.g. healthy, unhealthy) implicitly learn pathologies? Moreover, we investigate the effect of pretraining and data imbalance on the interpretability of learned features. In addition to the analysis, we propose semantic attribution to semantically explain each prediction. We present our findings using publicly available chest pathologies (CheXpert [5], NIH ChestX-ray8 [25]) and COVID-19 datasets (BrixIA [20], and COVID-19 chest X-ray segmentation dataset [4]).
Neural networks have demonstrated remarkable performance in classification and regression tasks on chest X-rays. In order to establish trust in the clinical routine, the networks’ prediction mechanism needs to be interpretable. One principal approach to interpretation is feature attribution. Feature attribution methods identify the importance of input features for the output prediction. Building on Information Bottleneck Attribution (IBA) method, for each prediction we identify the chest X-ray regions that have high mutual information with the network’s output. Original IBA identifies input regions that have sufficient predictive information. We propose Inverse IBA to identify all informative regions. Thus all predictive cues for pathologies are highlighted on the X-rays, a desirable property for chest X-ray diagnosis. Moreover, we propose Regression IBA for explaining regression models. Using Regression IBA we observe that a model trained on cumulative severity score labels implicitly learns the severity of different X-ray regions. Finally, we propose Multi-layer IBA to generate higher resolution and more detailed attribution/saliency maps. We evaluate our methods using both human-centric (ground-truth-based) interpretability metrics, and human-agnostic feature importance metrics on NIH Chest X-ray8 and BrixIA datasets.
Is critical input information encoded in specific sparse pathways within the neural network? In this work, we discuss the problem of identifying these critical pathways and subsequently leverage them for interpreting the network’s response to an input. The pruning objective — selecting the smallest group of neurons for which the response remains equivalent to the original network — has been previously proposed for identifying critical pathways. We demonstrate that sparse pathways derived from pruning do not necessarily encode critical input information. To ensure sparse pathways include critical fragments of the encoded input information, we propose pathway selection via neurons’ contribution to the response. We proceed to explain how critical pathways can reveal critical input features. We prove that pathways selected via neuron contribution are locally linear (in an ℓ 2 -ball), a property that we use for proposing a feature attribution method: ‘pathway gradient’. We validate our interpretation method using mainstream evaluation experiments. The validation of pathway gradient interpretation method further confirms that selected pathways using neuron contributions correspond to critical input features. The code 1 2 is publicly available.
An outside-in system uses binocular stereo and a probabilistic sparse point cloud matcher to track objects with micrometre precision in real-time. Miniaturizing the system results in a markerless inside-out stereo method with improved rotational accuracy. Reducing the constraints, we reformulate marker-free monocular pose estimation as an action decision process where the next best pose is determined using a render-and-compare strategy. This allows instance agnostic pose estimation that generalizes to unseen objects. The methods are applied on a set of medical and industrial applications.
Segmentation of Multiple Sclerosis (MS) lesions in longitudinal brain MR scans is performed for monitoring the progression of MS lesions. We hypothesize that the spatio-temporal cues in longitudinal data can aid the segmentation algorithm. Therefore, we propose a multi-task learning approach by defining an auxiliary self-supervised task of deformable registration between two time-points to guide the neural network toward learning from spatio-temporal changes. We show the efficacy of our method on a clinical dataset comprised of 70 patients with one follow-up study for each patient. Our results show that spatio-temporal information in longitudinal data is a beneficial cue for improving segmentation. We improve the result of current state-of-the-art by 2.6% in terms of overall score (p < 0.05).
Federated learning (FL) has been a promising approach in the field of medical imaging in recent years. A critical problem in FL, specifically in medical scenarios is to have a more accurate shared model which is robust to noisy and out-of distribution clients. In this work, we tackle the problem of statistical heterogeneity in data for FL which is highly plausible in medical data where for example the data comes from different sites with different scanner settings. We propose IDA (Inverse Distance Aggregation), a novel adaptive weighting approach for clients based on meta-information which handles unbalanced and non-iid data. We extensively analyze and evaluate our method against the well-known FL approach, Federated Averaging as a baseline.
©all images: LMU | TUM