A5002648077- Radiomics and Machine Learning in Medical Imaging
- AI in cancer detection
- Brain Tumor Detection and Classification
- Medical Image Segmentation Techniques
- Artificial Intelligence in Healthcare and Education
- Viral Infectious Diseases and Gene Expression in Insects
- Adversarial Robustness in Machine Learning
- Advanced X-ray and CT Imaging
- Medical Imaging Techniques and Applications
- Medical Imaging and Analysis
- Healthcare Technology and Patient Monitoring
- Retinal Imaging and Analysis
- Cerebrovascular and Carotid Artery Diseases
- Additive Manufacturing and 3D Printing Technologies
- Modular Robots and Swarm Intelligence
- Radiology practices and education
- Risk and Safety Analysis
- Intracranial Aneurysms: Treatment and Complications
- Dental Radiography and Imaging
- Robotics and Sensor-Based Localization
- Manufacturing Process and Optimization
- Explainable Artificial Intelligence (XAI)
- Advanced Radiotherapy Techniques
- Machine Learning in Materials Science
- Radiation Dose and Imaging
Heidelberg University
2023-2024
German Cancer Research Center
2023-2024
Foundation models have taken over natural language processing and image generation domains due to the flexibility of prompting. With recent introduction Segment Anything Model (SAM), this prompt-driven paradigm has entered segmentation with a hitherto unexplored abundance capabilities. The purpose paper is conduct an initial evaluation out-of-the-box zero-shot capabilities SAM for medical segmentation, by evaluating its performance on abdominal CT organ task, via point or bounding box based...
Breast cancer is one of the most common causes death among women worldwide. Early detection helps in reducing number deaths. Automated 3D Ultrasound (ABUS) a newer approach for breast screening, which has many advantages over handheld mammography such as safety, speed, and higher rate cancer. Tumor detection, segmentation, classification are key components analysis medical images, especially challenging context ABUS due to significant variability tumor size shape, unclear boundaries, low...
Building trusted datasets is critical for transparent and responsible Medical AI (MAI) research, but creating even small, high-quality can take years of effort from multidisciplinary teams. This process often delays benefits, as human-centric data creation AI-centric model development are treated separate, sequential steps. To overcome this, we propose ScaleMAI, an agent AI-integrated curation annotation, allowing quality performance to improve in a self-reinforcing cycle reducing time...
Multi-class segmentation of the aorta in computed tomography angiography (CTA) scans is essential for diagnosing and planning complex endovascular treatments patients with aortic dissections. However, existing methods reduce to a binary problem, limiting their ability measure diameters across different branches zones. Furthermore, no open-source dataset currently available support development multi-class methods. To address this gap, we organized AortaSeg24 MICCAI Challenge, introducing...
The Circle of Willis (CoW) is an important network arteries connecting major circulations the brain. Its vascular architecture believed to affect risk, severity, and clinical outcome serious neuro-vascular diseases. However, characterizing highly variable CoW anatomy still a manual time-consuming expert task. usually imaged by two angiographic imaging modalities, magnetic resonance angiography (MRA) computed tomography (CTA), but there exist limited public datasets with annotations on...
Accurately segmenting thin tubular structures, such as vessels, nerves, roads or concrete cracks, is a crucial task in computer vision. Standard deep learning-based segmentation loss functions, Dice Cross-Entropy, focus on volumetric overlap, often at the expense of preserving structural connectivity topology. This can lead to errors that adversely affect downstream tasks, including flow calculation, navigation, and inspection. Although current topology-focused losses mark an improvement,...
Despite considerable strides in developing deep learning models for 3D medical image segmentation, the challenge of effectively generalizing across diverse distributions persists. While domain generalization is acknowledged as vital robust application clinical settings, challenges stemming from training with a limited Field View (FOV) remain unaddressed. This limitation leads to false predictions when applied body regions beyond FOV data. In response this problem, we propose novel loss...
Automated lesion segmentation in PET/CT scans is crucial for improving clinical workflows and advancing cancer diagnostics. However, the task challenging due to physiological variability, different tracers used PET imaging, diverse imaging protocols across medical centers. To address this, autoPET series was created challenge researchers develop algorithms that generalize environments. This paper presents our solution III challenge, targeting multitracer, multicenter generalization using...
The third autoPET challenge introduced a new data-centric task this year, shifting the focus from model development to improving metastatic lesion segmentation on PET/CT images through data quality and handling strategies. In response, we developed targeted methods enhance performance tailored characteristics of imaging. Our approach encompasses two key elements. First, address potential alignment errors between CT PET modalities as well prevalence punctate lesions, modified baseline...
Accurate segmentation of Multiple Sclerosis (MS) lesions in longitudinal MRI scans is crucial for monitoring disease progression and treatment efficacy. Although changes across time are taken into account when assessing images clinical practice, most existing deep learning methods treat from different timepoints separately. Among studies utilizing images, a simple channel-wise concatenation the primary albeit suboptimal method employed to integrate timepoints. We introduce novel approach...
How can we test AI performance? This question seems trivial, but it isn't. Standard benchmarks often have problems such as in-distribution and small-size sets, oversimplified metrics, unfair comparisons, short-term outcome pressure. As a consequence, good performance on standard does not guarantee success in real-world scenarios. To address these problems, present Touchstone, large-scale collaborative segmentation benchmark of 9 types abdominal organs. is based 5,195 training CT scans from...
Current interactive segmentation approaches, inspired by the success of META's Segment Anything model, have achieved notable advancements, however, they come with substantial limitations that hinder their practical application in real clinical scenarios. These include unrealistic human interaction requirements, such as slice-by-slice operations for 2D models on 3D data, a lack iterative refinement, and insufficient evaluation experiments. shortcomings prevent accurate assessment model...
Motivation: Tensor-encoded diffusion MRI (dMRI) methods for tissue microstructure elucidation typically require lengthy dMRI acquisitions and computationally costly, SNR-sensitive data analysis. Goal(s): Employing q-space trajectory imaging (QTI), we seek to greatly reduce both the number of required measurements computational burden in analysis robust estimation parameters quantifying brain microstructure. Approach: A machine learning-based estimator is trained on a 10-fold reduced subset...
This paper presents our approach to scaling the nnU-Net framework for multi-structure segmentation on Cone Beam Computed Tomography (CBCT) images, specifically in scope of ToothFairy2 Challenge. We leveraged ResEnc L model, introducing key modifications patch size, network topology, and data augmentation strategies address unique challenges dental CBCT imaging. Our method achieved a mean Dice coefficient 0.9253 HD95 18.472 test set, securing rank 4.6 with it first place challenge. The source...
In recent years, several algorithms have been developed for the segmentation of Inferior Alveolar Canal (IAC) in Cone-Beam Computed Tomography (CBCT) scans. However, availability public datasets this domain is limited, resulting a lack comparative evaluation studies on common benchmark. To address scientific gap and encourage deep learning research field, ToothFairy challenge was organized within MICCAI 2023 conference. context, dataset released to also serve as benchmark future research....