Background: – Assessment of pathology endpoints in animal models of diabetic kidney disease (DKD) is time-consuming and prone to expert bias. Additionally, the sparsity of human kidney biopsy data hinders the development of translational models from animals to humans.Methods: – We developed an AI-driven workflow to streamline histopathological assessments in animal models of diabetic nephropathy. Our approach (i) detected glomeruli in whole slide images, (ii) enabled fast expert scoring via an annotation tool, and (iii) automated scoring. By leveraging unlabeled preclinical data for self-supervised learning, we enhanced AI scoring performance, reduced expert bias, and enabled the translation of AI scoring from animal models to human biopsies. To translate AI models from preclinical studies to human biopsies, we introduced a method that adjusted the feature extractor to human-specific features during inference without the need for annotated examples.Results: – Our annotation tool streamlined glomerular scoring, reducing turnaround time by 80%. Supervised AI models outperformed expert agreement and further reduced turnaround time by 90%, demonstrating generalization across studies involving both the same and different animal models. Without supervision, the self-supervised model achieved a κ value of 0.78, effectively identifying glomerular changes without guidance. Incorporating self-supervised learning into supervised training improved performance to κ = 0.84 and reduced bias compared to individual experts (P < 0.001). Our translational approach achieved a κ value of 0.63 on human glomeruli, even though the model was trained exclusively on mouse glomeruli scores, reducing the translational gap by 45%.Conclusions: – In this study, we accelerated and enhanced pathology readouts in a real-life pharmaceutical industry setting. We show that AI-assisted scoring reduced pathologists' workload and expedited study assessments. Self-supervised learning captured intrinsic properties of kidney morphology without expert annotation, reduced expert bias and translational discrepancies, greatly facilitating translational activities in drug development for patients with DKD.

Texture recognition has recently been dominated by ImageNet-pre-trained deep Convolutional Neural Networks (CNNs), with specialized modifications and feature engineering required to achieve state-of-the-art (SOTA) per formance. However, although Vision Transformers (ViTs) were introduced a few years ago, little is known about their texture recognition abilities. Therefore, in this work, we introduce VORTEX (ViTs with Orderless and Randomized Token Encodings for Texture Recognition), a novel method that enables the effective use of ViTs for texture analysis. VORTEX extracts multi-depth token embeddings from pre-trained ViT backbones and em ploys a lightweight module to aggregate hierarchical features and perform orderless encoding, obtaining a better image representation for texture recognition tasks. This approach allows seamless integration with any ViT us ing the common transformer architecture. Moreover, no fine-tuning of the backbone is performed, as they are used only as frozen feature extractors, and the features are fed to a linear SVM. We evaluate VORTEX on nine diverse texture datasets, demonstrating its ability to achieve or surpass SOTA performance in a variety of texture analysis scenarios. By bridging the gap between texture recognition with CNNs and transformer-based archi tectures, VORTEX paves the way for adopting emerging transformer foundation models. Furthermore, VORTEX demonstrates robust computational efficiency when coupled with ViT backbones compared to CNNs with sim ilar costs. The method implementation and experimental scripts are publicly available in our online repository (https://github.com/scabini/VORTEX).

Self-supervised adaptation (SSA) improves foundation model transfer to medical domains but is computationally prohibitive. Although parameter efficient fine-tuning (PEFT) methods such as LoRA have been explored for supervised adaptation, their effectiveness for SSA remains unknown. In this work, we introduce efficient self-supervised adaptation (ESSA), a framework that applies parameter-efficient fine-tuning techniques to SSA with the aim of reducing computational cost and improving adaptation performance. To the best of our knowledge, we are the first to demonstrate that PEFT methods can be effectively applied to SSA to improve self-supervised learning, challenging the assumption that full-parameter SSA is necessary for optimal performance. Furthermore, we show that applying PEFT during supervised adaptation following self-supervision leads to additional performance gains, outperforming full-parameter training.

Ovarian lesions are common and often incidentally detected. A critical shortage of expert ultrasound examiners has raised concerns of unnecessary interventions and delayed cancer diagnoses. Deep learning has shown promising results in the detection of ovarian cancer in ultrasound images; however, external validation is lacking. In this international multicenter retrospective study, we developed and validated transformer-based neural network models using a comprehensive dataset of 17,119 ultrasound images from 3,652 patients across 20 centers in eight countries. Using a leave-one-center-out cross-validation scheme, for each center in turn, we trained a model using data from the remaining centers. The models demonstrated robust performance across centers, ultrasound systems, histological diagnoses and patient age groups, significantly outperforming both expert and non-expert examiners on all evaluated metrics, namely F1 score, sensitivity, specificity, accuracy, Cohen’s kappa, Matthew’s correlation coefficient, diagnostic odds ratio and Youden’s J statistic. Furthermore, in a retrospective triage simulation, artificial intelligence (AI)-driven diagnostic support reduced referrals to experts by 63% while significantly surpassing the diagnostic performance of the current practice. These results show that transformer-based models exhibit strong generalization and above human expert-level diagnostic accuracy, with the potential to alleviate the shortage of expert ultrasound examiners and improve patient outcomes.

How well can one expect transfer learning to work in a new setting where the domain is shifted, the task is different, and the architecture changes? Many transfer learning metrics have been proposed to answer this question. But how accurate are their predictions in a realistic new setting? We conducted an extensive evaluation involving over 42,000 experiments comparing 23 transferability metrics across 16 different datasets to assess their ability to predict transfer performance for image classification tasks. Our findings reveal that none of the existing metrics perform well across the board. However, we find that a simple k-nearest neighbor evaluation – as is commonly used to evaluate feature quality for self-supervision – not only surpasses existing metrics, but also offers better computational efficiency and ease of implementation.

We investigate the role of uncertainty quantification in aiding medical decision-making. Existing evaluation metrics fail to capture the practical utility of joint human-AI decision-making systems. To address this, we introduce a novel framework to assess such systems and use it to benchmark a diverse set of confidence and uncertainty estimation methods. Our results show that certainty measures enable joint human-AI systems to outperform both standalone humans and AIs, and that for a given system there exists an optimal balance in the number of cases to refer to humans, beyond which the system’s performance degrades.

We investigate the role of uncertainty quantification in aiding medical decision-making. Existing evaluation metrics fail to capture the practical utility of joint human-AI decision-making systems. To address this, we introduce a novel framework to assess such systems and use it to benchmark a diverse set of confidence and uncertainty estimation methods. Our results show that certainty measures enable joint human-AI systems to outperform both standalone humans and AIs, and that for a given system there exists an optimal balance in the number of cases to refer to humans, beyond which the system's performance degrades.

Screening mammography reduces breast cancer mortality, but studies analyzing interval cancers diagnosed after negative screens have shown that many cancers are missed. Supplemental screening using magnetic resonance imaging (MRI) can reduce the number of missed cancers. However, as qualified MRI staff are lacking, the equipment is expensive to purchase and cost-effectiveness for screening may not be convincing, the utilization of MRI is currently limited. An effective method for triaging individuals to supplemental MRI screening is therefore needed. We conducted a randomized clinical trial, ScreenTrustMRI, using a recently developed artificial intelligence (AI) tool to score each mammogram. We offered trial participation to individuals with a negative screening mammogram and a high AI score (top 6.9%). Upon agreeing to participate, individuals were assigned randomly to one of two groups: those receiving supplemental MRI and those not receiving MRI. The primary endpoint of ScreenTrustMRI is advanced breast cancer defined as either interval cancer, invasive component larger than 15 mm or lymph node positive cancer, based on a 27-month follow-up time from the initial screening. Secondary endpoints, prespecified in the study protocol to be reported before the primary outcome, include cancer detected by supplemental MRI, which is the focus of the current paper. Compared with traditional breast density measures used in a previous clinical trial, the current AI method was nearly four times more efficient in terms of cancers detected per 1,000 MRI examinations (64 versus 16.5). Most additional cancers detected were invasive and several were multifocal, suggesting that their detection was timely. Altogether, our results show that using an AI-based score to select a small proportion (6.9%) of individuals for supplemental MRI after negative mammography detects many missed cancers, making the cost per cancer detected comparable with screening mammography. ClinicalTrials.gov registration: NCT04832594.

The deep learning field is converging towards the use of general foundation models that can be easily adapted for diverse tasks. While this paradigm shift has become common practice within the field of natural language processing, progress has been slower in computer vision. In this paper we attempt to address this issue by investigating the transferability of various state-of-the-art foundation models to medical image classification tasks. Specifically, we evaluate the performance of five foundation models, namely Sam, Seem, Dinov2, BLIP, and OpenCLIP across four well-established medical imaging datasets. We explore different training settings to fully harness the potential of these models. Our study shows mixed results. Dinov2 consistently outperforms the standard practice of ImageNet pretraining. However, other foundation models failed to consistently beat this established baseline indicating limitations in their transferability to medical image classification tasks.

High Content Imaging (HCI) plays a vital role in modern drug discovery and development pipelines, facilitating various stages from hit identification to candidate drug characterization. Applying machine learning models to these datasets can prove challenging as they typically consist of multiple batches, affected by experimental variation, especially if different imaging equipment have been used. Moreover, as new data arrive, it is preferable that they are analyzed in an online fashion. To overcome this, we propose CODA, an online self-supervised domain adaptation approach. CODA divides the classifier’s role into a generic feature extractor and a task-specific model. We adapt the feature extractor’s weights to the new domain using cross-batch self-supervision while keeping the task-specific model unchanged. Our results demonstrate that this strategy significantly reduces the generalization gap, achieving up to a 300% improvement when applied to data from different labs utilizing different microscopes. CODA can be applied to new, unlabeled out-of-domain data sources of different sizes, from a single plate to multiple experimental batches.