With access to large-scale, unlabeled medical datasets, researchers are confronted with two questions: Should they attempt to pretrain a custom foundation model on this medical data, or use transfer-learning from an existing generalist model? And, if a custom model is pretrained, are novel methods required? In this paper we explore these questions by conducting a case-study, in which we train a foundation model on a large regional fetal ultrasound dataset of 2M images. By selecting the well-established DINOv2 method for pretraining, we achieve state-of-the-art results on three fetal ultrasound datasets, covering data from different countries, classification, segmentation, and few-shot tasks. We compare against a series of models pretrained on natural images, ultrasound images, and supervised baselines. Our results demonstrate two key insights: (i) Pretraining on custom data is worth it, even if smaller models are trained on less data, as scaling in natural image pretraining does not translate to ultrasound performance. (ii) Well-tuned methods from computer vision are making it feasible to train custom foundation models for a given medical domain, requiring no hyperparameter tuning and little methodological adaptation. Given these findings, we argue that a bias towards methodological innovation should be avoided when developing domain specific foundation models under common computational resource constraints.

Early detection through screening is critical for reducing gastric cancer (GC) mortality. However, in most high-prevalence regions, large-scale screening remains challenging due to limited resources, low compliance and suboptimal detection rate of upper endoscopic screening. Therefore, there is an urgent need for more efficient screening protocols. Noncontrast computed tomography (CT), routinely performed for clinical purposes, presents a promising avenue for large-scale designed or opportunistic screening. Here we developed the Gastric Cancer Risk Assessment Procedure with Artificial Intelligence (GRAPE), leveraging noncontrast CT and deep learning to identify GC. Our study comprised three phases. First, we developed GRAPE using a cohort from 2 centers in China (3,470 GC and 3,250 non-GC cases) and validated its performance on an internal validation set (1,298 cases, area under curve = 0.970) and an independent external cohort from 16 centers (18,160 cases, area under curve = 0.927). Subgroup analysis showed that the detection rate of GRAPE increased with advancing T stage but was independent of tumor location. Next, we compared the interpretations of GRAPE with those of radiologists and assessed its potential in assisting diagnostic interpretation. Reader studies demonstrated that GRAPE significantly outperformed radiologists, improving sensitivity by 21.8% and specificity by 14.0%, particularly in early-stage GC. Finally, we evaluated GRAPE in real-world opportunistic screening using 78,593 consecutive noncontrast CT scans from a comprehensive cancer center and 2 independent regional hospitals. GRAPE identified persons at high risk with GC detection rates of 24.5% and 17.7% in 2 regional hospitals, with 23.2% and 26.8% of detected cases in T1/T2 stage. Additionally, GRAPE detected GC cases that radiologists had initially missed, enabling earlier diagnosis of GC during follow-up for other diseases. In conclusion, GRAPE demonstrates strong potential for large-scale GC screening, offering a feasible and effective approach for early detection. ClinicalTrials.gov registration: NCT06614179 .

This study aimed to develop and validate a multimodal deep learning model that leverages 2D grayscale ultrasound (US) images alongside readily available clinical data to improve diagnostic performance for ovarian cancer (OC). A retrospective analysis was conducted involving 1899 patients who underwent preoperative US examinations and subsequent surgeries for adnexal masses between 2019 and 2024. A multimodal deep learning model was constructed for OC diagnosis and extracting US morphological features from the images. The model's performance was evaluated using metrics such as receiver operating characteristic (ROC) curves, accuracy, and F1 score. The multimodal deep learning model exhibited superior performance compared to the image-only model, achieving areas under the curves (AUCs) of 0.9393 (95% CI 0.9139-0.9648) and 0.9317 (95% CI 0.9062-0.9573) in the internal and external test sets, respectively. The model significantly improved the AUCs for OC diagnosis by radiologists and enhanced inter-reader agreement. Regarding US morphological feature extraction, the model demonstrated robust performance, attaining accuracies of 86.34% and 85.62% in the internal and external test sets, respectively. Multimodal deep learning has the potential to enhance the diagnostic accuracy and consistency of radiologists in identifying OC. The model's effective feature extraction from ultrasound images underscores the capability of multimodal deep learning to automate the generation of structured ultrasound reports.

Panoramic X-rays (PX) are extensively utilized in dental and maxillofacial diagnostics, offering comprehensive imaging of teeth and surrounding structures. This study investigates the automatic determination of Kennedy's classification in partially edentulous jaws. A retrospective study involving 209 PX images from 206 patients was conducted. The established Mask R-CNN, a deep learning-based instance segmentation model, was trained for the automatic detection, position labeling (according to the international dental federation's scheme), and segmentation of teeth in PX. Subsequent post-processing steps filter duplicate outputs by position label and by geometric overlap. Finally, a rule-based determination of Kennedy's class of partially edentulous jaws was performed. In a fivefold cross-validation, Kennedy's classification was correctly determined in 83.0% of cases, with the most common errors arising from the mislabeling of morphologically similar teeth. The underlying algorithm demonstrated high sensitivity (97.1%) and precision (98.1%) in tooth detection, with an F1 score of 97.6%. FDI position label accuracy was 94.7%. Ablation studies indicated that post-processing steps, such as duplicate filtering, significantly improved algorithm performance. Our findings show that automatic dentition analysis in PX images can be extended to include clinically relevant jaw classification, reducing the workload associated with manual labeling and classification.

The success of deep learning in medical imaging applications has led several companies to deploy proprietary models in diagnostic workflows, offering monetized services. Even though model weights are hidden to protect the intellectual property of the service provider, these models are exposed to model stealing (MS) attacks, where adversaries can clone the model's functionality by querying it with a proxy dataset and training a thief model on the acquired predictions. While extensively studied on general vision tasks, the susceptibility of medical imaging models to MS attacks remains inadequately explored. This paper investigates the vulnerability of black-box medical imaging models to MS attacks under realistic conditions where the adversary lacks access to the victim model's training data and operates with limited query budgets. We demonstrate that adversaries can effectively execute MS attacks by using publicly available datasets. To further enhance MS capabilities with limited query budgets, we propose a two-step model stealing approach termed QueryWise. This method capitalizes on unlabeled data obtained from a proxy distribution to train the thief model without incurring additional queries. Evaluation on two medical imaging models for Gallbladder Cancer and COVID-19 classification substantiates the effectiveness of the proposed attack. The source code is available at https://github.com/rajankita/QueryWise.

Artifacts in kilo-Voltage CT (kVCT) imaging degrade image quality, impacting clinical decisions. We propose a deep learning framework for metal artifact reduction (MAR) and domain transformation from kVCT to Mega-Voltage CT (MVCT). The proposed framework, ReMAR-DS, utilizes an encoder-decoder architecture with enhanced feature recalibration, effectively reducing artifacts while preserving anatomical structures. This ensures that only relevant information is utilized in the reconstruction process. By infusing recalibrated features from the encoder block, the model focuses on relevant spatial regions (e.g., areas with artifacts) and highlights key features across channels (e.g., anatomical structures), leading to improved reconstruction of artifact-corrupted regions. Unlike traditional MAR methods, our approach bridges the gap between high-resolution kVCT and artifact-resistant MVCT, enhancing radiotherapy planning. It produces high-quality MVCT-like reconstructions, validated through qualitative and quantitative evaluations. Clinically, this enables oncologists to rely on kVCT alone, reducing repeated high-dose MVCT scans and lowering radiation exposure for cancer patients.

Vascular diseases pose a significant threat to human health, with X-ray angiography established as the gold standard for diagnosis, allowing for detailed observation of blood vessels. However, angiographic X-rays expose personnel and patients to higher radiation levels than non-angiographic X-rays, which are unwanted. Thus, modality translation from non-angiographic to angiographic X-rays is desirable. Data-driven deep approaches are hindered by the lack of paired large-scale X-ray angiography datasets. While making high-quality vascular angiography synthesis crucial, it remains challenging. We find that current medical image synthesis primarily operates at pixel level and struggles to adapt to the complex geometric structure of blood vessels, resulting in unsatisfactory quality of blood vessel image synthesis, such as disconnections or unnatural curvatures. To overcome this issue, we propose a self-supervised method via diffusion models to transform non-angiographic X-rays into angiographic X-rays, mitigating data shortages for data-driven approaches. Our model comprises a diffusion model that learns the distribution of vascular data from diffusion latent, a generator for vessel synthesis, and a mask-based adversarial module. To enhance geometric accuracy, we propose a parametric vascular model to fit the shape and distribution of blood vessels. The proposed method contributes a pipeline and a synthetic dataset for X-ray angiography. We conducted extensive comparative and ablation experiments to evaluate the Angio-Diff. The results demonstrate that our method achieves state-of-the-art performance in synthetic angiography image quality and more accurately synthesizes the geometric structure of blood vessels. The code is available at https://github.com/zfw-cv/AngioDiff.

Regular mammography screening is essential for early breast cancer detection. Deep learning-based risk prediction methods have sparked interest to adjust screening intervals for high-risk groups. While early methods focused only on current mammograms, recent approaches leverage the temporal aspect of screenings to track breast tissue changes over time, requiring spatial alignment across different time points. Two main strategies for this have emerged: explicit feature alignment through deformable registration and implicit learned alignment using techniques like transformers, with the former providing more control. However, the optimal approach for explicit alignment in mammography remains underexplored. In this study, we provide insights into where explicit alignment should occur (input space vs. representation space) and if alignment and risk prediction should be jointly optimized. We demonstrate that jointly learning explicit alignment in representation space while optimizing risk estimation performance, as done in the current state-of-the-art approach, results in a trade-off between alignment quality and predictive performance and show that image-level alignment is superior to representation-level alignment, leading to better deformation field quality and enhanced risk prediction accuracy. The code is available at https://github.com/sot176/Longitudinal_Mammogram_Alignment.git.

Convolutional denoising autoencoders (DAEs) are powerful tools for image restoration. However, they inherit a key limitation of convolutional neural networks (CNNs): they tend to recover low-frequency features, such as smooth regions, more effectively than high-frequency details. This leads to the loss of fine details, which is particularly problematic in dental radiographs where preserving subtle anatomical structures is crucial. While self-attention mechanisms can help mitigate this issue by emphasizing important features, conventional attention methods often prioritize features corresponding to cleaner regions and may overlook those obscured by noise. To address this limitation, we propose a noise-aware self-attention method, which allows the model to effectively focus on and recover key features even within noisy regions. Building on this approach, we introduce the noise-aware attention-enhanced denoising autoencoder (NAADA) network for enhancing noisy panoramic dental radiographs. Compared with the recent state of the art (and much heavier) methods like Uformer, MResDNN etc., our method improves the reconstruction of fine details, ensuring better image quality and diagnostic accuracy.

Computed Tomography (CT) plays a crucial role in clinical diagnosis, but the growing demand for CT examinations has raised concerns about diagnostic errors. While Multimodal Large Language Models (MLLMs) demonstrate promising comprehension of medical knowledge, their tendency to produce inaccurate information highlights the need for rigorous validation. However, existing medical visual question answering (VQA) benchmarks primarily focus on simple visual recognition tasks, lacking clinical relevance and failing to assess expert-level knowledge. We introduce MedErr-CT, a novel benchmark for evaluating medical MLLMs' ability to identify and correct errors in CT reports through a VQA framework. The benchmark includes six error categories - four vision-centric errors (Omission, Insertion, Direction, Size) and two lexical error types (Unit, Typo) - and is organized into three task levels: classification, detection, and correction. Using this benchmark, we quantitatively assess the performance of state-of-the-art 3D medical MLLMs, revealing substantial variation in their capabilities across different error types. Our benchmark contributes to the development of more reliable and clinically applicable MLLMs, ultimately helping reduce diagnostic errors and improve accuracy in clinical practice. The code and datasets are available at https://github.com/babbu3682/MedErr-CT.