Towards Improved Clinical Adoption of AI Segmentation Models: Benchmarking High-Performance Models for Resource-Constrained Settings

Foundation models are pretrained on massive datasets. However, collecting medical datasets is expensive and time-consuming, and raises privacy concerns. Here we show that synthetic data generated via conditioning with disease labels can be leveraged for building high-performing medical foundation models. We pretrained a retinal foundation model, first with approximately one million synthetic retinal images with physiological structures and feature distribution consistent with real counterparts, and then with only 16.7% of the 904,170 real-world colour fundus photography images required in a recently reported retinal foundation model (RETFound). The data-efficient model performed as well or better than RETFound across nine public datasets and four diagnostic tasks; and for diabetic-retinopathy grading, it used only 40% of the expert-annotated training data used by RETFound. We also support the generalizability of the data-efficient strategy by building a classifier for the detection of tuberculosis on chest X-ray images. The text-conditioned generation of synthetic data may enhance the performance and generalization of medical foundation models.

While fully supervised learning can yield high-performing segmentation models, the effort required to manually segment large training sets limits practical utility. We investigate whether data mined line annotations can facilitate brain MRI tumor segmentation model development without requiring manually segmented training data. In this retrospective study, a tumor detection model trained using clinical line annotations mined from PACS was leveraged with unsupervised segmentation to generate pseudo-masks of enhancing tumors on T1-weighted post-contrast images (9911 image slices; 3449 adult patients). Baseline segmentation models were trained and employed within a semi-supervised learning (SSL) framework to refine the pseudo-masks. Following each self-refinement cycle, a new model was trained and tested on a held-out set of 319 manually segmented image slices (93 adult patients), with the SSL cycles continuing until Dice score coefficient (DSC) peaked. DSCs were compared using bootstrap resampling. Utilizing the best-performing models, two inference methods were compared: (1) conventional full-image segmentation, and (2) a hybrid method augmenting full-image segmentation with detection plus image patch segmentation. Baseline segmentation models achieved DSC of 0.768 (U-Net), 0.831 (Mask R-CNN), and 0.838 (HRNet), improving with self-refinement to 0.798, 0.871, and 0.873 (each p < 0.001), respectively. Hybrid inference outperformed full image segmentation alone: DSC 0.884 (Mask R-CNN) vs. 0.873 (HRNet), p < 0.001. Line annotations mined from PACS can be harnessed within an automated pipeline to produce accurate brain MRI tumor segmentation models without manually segmented training data, providing a mechanism to rapidly establish tumor segmentation capabilities across radiology modalities. • A brain MRI tumor detection model trained using clinical line measurement annotations mined from PACS was leveraged to automatically generate tumor segmentation pseudo-masks. • An iterative self-refinement process automatically improved pseudo-mask quality, with the best-performing segmentation pipeline achieving a Dice score of 0.884 on a held-out test set. • Tumor line measurement annotations generated in routine clinical radiology practice can be harnessed to develop high-performing segmentation models without manually segmented training data, providing a mechanism to rapidly establish tumor segmentation capabilities across radiology modalities.

There are two largely competing models for an athletics organization at both the collegiate and professional levels: the High Performance Model and the Medical Model. The High Performance Model arises from international Football perspective that places a "Performance Director" at the center of teams supporting the athletes. The Medical Model, supported by both the National Athletic Trainers Association and the NCAA, separates off medical staff (athletic trainers, physical therapists, and physicians, predominantly) and emphasizes the autonomy of medical decisions. The Medical Model has left clinicians in a "medical silo", limiting our ability to care for the individual athletes as holistic people and limits our wider impact in the field of athlete health and injury mitigation. We argue that Medical. Model is consistent with the High Performance Model only if we reject the notion that the "Performance Director" is an administrative person and instead conceptualize this as a "Health and Performance Information Hub" which facilitates transdisciplinary collaboration. This Commentary details how a data broker system can be used to accelerate transdisciplinary collaboration within an athletic organization, leading to better healthcare for athletes and improved team and individual performance. Furthermore, a transdisciplinary organization with data sharing is able to turn real-world data into real-world evidence, enhancing the care and performance of athletes locally as well as facilitating the creation of generalizable knowledge in the area of sports medicine and human performance.

This study investigates the comparative performance of foundational models, advanced large-kernel architectures, and traditional deep learning approaches for hyperpolarized gas MRI segmentation across progressive data reduction scenarios. Chronic obstructive pulmonary disease (COPD) remains a leading global health concern, and advanced imaging techniques are crucial for its diagnosis and management. Hyperpolarized gas MRI, utilizing helium-3 (3He) and xenon-129 (129Xe), offers a non-invasive way to assess lung function. We evaluated foundational models (Segment Anything Model and MedSAM), advanced architectures (UniRepLKNet and TransXNet), and traditional deep learning models (UNet with VGG19 backbone, Feature Pyramid Network with MIT-B5 backbone, and DeepLabV3 with ResNet152 backbone) using four data availability scenarios: 100%, 50%, 25%, and 10% of the full training dataset (1640 2D MRI slices from 205 participants). The results demonstrate that foundational and advanced models achieve statistically equivalent performance across all data scenarios (p > 0.01), while both significantly outperform traditional architectures under data constraints (p < 0.001). Under extreme data scarcity (10% training data), foundational and advanced models maintained DSC values above 0.86, while traditional models experienced catastrophic performance collapse. This work highlights the critical advantage of architectures with large effective receptive fields in medical imaging applications where data collection is challenging, demonstrating their potential to democratize advanced medical imaging analysis in resource-limited settings.

Learning from multi-center medical datasets to obtain a high-performance global model is challenging due to the privacy protection and data heterogeneity in healthcare systems. Current federated learning approaches are not efficient enough to learn Non-Independent and Identically Distributed (Non-IID) data and require high communication costs. In this work, a practical privacy computing framework is proposed to train a Non-IID medical image segmentation model under various multi-center setting in low communication cost. Specifically, an efficient cascaded diffusion model is trained to generate image-mask pairs that have similar distribution to the training data of clients, providing rich labeled data on client side to mitigate heterogeneity. Also, a label construction module is developed to improve the quality of generated image-mask pairs. Moreover, a set of aggregation methods is proposed to achieve global model from data generated from Cascaded Diffusion model for diverse scenarios: CD-Syn, CD-Ens and its extension CD-KD. CD-Syn is a one-shot method that trains segmentation model solely on public generated datasets while CD-Ens and CD-KD maximize the utilization of local original data by an extra communication round of ensemble or knowledge distillation. In this way, the setting of our proposed framework is highly practical, providing multiple aggregation methods which can flexibly adapt to varying demands for efficiency, privacy, and accuracy. We systematically evaluated the effectiveness of our proposed framework on five Non-IID medical datasets and observe 5.38% improvement in Dice score compared with baseline method (FednnU-Net) on average.

BrainSegFounder: Towards 3D foundation models for neuroimage segmentation

Efficient and lightweight layer-wise in-situ defect detection in laser powder bed fusion via knowledge distillation and structural re-parameterization

Prostate cancer (PCa) is one of the most common types of cancer, seriously affecting adult male health. Accurate and automated PCa segmentation is essential for radiologists to confirm the location of cancer, evaluate its severity, and design appropriate treatments. This paper presents PCaSAM, a fully automated PCa segmentation model that allows us to input multi-modal MRI images into the foundation model to improve performance significantly. We collected multi-center datasets to conduct a comprehensive evaluation. The results showed that PCaSAM outperforms the generalist medical foundation model and the other representative segmentation models, with the average DSC of 0.721 and 0.706 in the internal and external datasets, respectively. Furthermore, with the assistance of segmentation, the PI-RADS scoring of PCa lesions was improved significantly, leading to a substantial increase in average AUC by 8.3–8.9% on two external datasets. Besides, PCaSAM achieved superior efficiency, making it highly suitable for real-world deployment scenarios.

PyMIC: A deep learning toolkit for annotation-efficient medical image segmentation

Deep neural networks show great potential as solutions to many sensing application problems, but their excessive resource demand slows down execution time, pausing a serious impediment to deployment on low-end devices. To address this challenge, recent literature focused on compressing neural network size to improve performance. We show that changing neural network size does not proportionally affect performance attributes of interest, such as execution time. Rather, extreme run-time nonlinearities exist over the network configuration space. Hence, we propose a novel framework, called FastDeepIoT, that uncovers the non-linear relation between neural network structure and execution time, then exploits that understanding to find network configurations that significantly improve the trade-off between execution time and accuracy on mobile and embedded devices. FastDeepIoT makes two key contributions. First, FastDeepIoT automatically learns an accurate and highly interpretable execution time model for deep neural networks on the target device. This is done without prior knowledge of either the hardware specifications or the detailed implementation of the used deep learning library. Second, FastDeepIoT informs a compression algorithm how to minimize execution time on the profiled device without impacting accuracy. We evaluate FastDeepIoT using three different sensing-related tasks on two mobile devices: Nexus 5 and Galaxy Nexus. FastDeepIoT further reduces the neural network execution time by $48\%$ to $78\%$ and energy consumption by $37\%$ to $69\%$ compared with the state-of-the-art compression algorithms.

With the development of deep learning in the field of medical image segmentation, various network segmentation models have been developed. Currently, the most common network models in medical image segmentation can be roughly categorized into pure convolutional networks, Transformer-based networks, and networks combining convolution and Transformer architectures. However, when dealing with complex variations and irregular shapes in medical images, existing networks face issues such as incomplete information extraction, large model parameter sizes, high computational complexity, and long processing times. In contrast, models with lower parameter counts and complexity can efficiently, quickly, and accurately identify lesion areas, significantly reducing diagnosis time and providing valuable time for subsequent treatments. Therefore, this paper proposes a lightweight network named MCI-Net, with only 5.48 M parameters, a computational complexity of 4.41, and a time complexity of just 0.263. By performing linear modeling on sequences, MCI-Net permanently marks effective features and filters out irrelevant information. It efficiently captures local-global information with a small number of channels, reduces the number of parameters, and utilizes attention calculations with exchange value mapping. This achieves model lightweighting and enables thorough interaction of local-global information within the computation, establishing an overall semantic relationship of local-global information. To verify the effectiveness of the MCI-Net network, we conducted comparative experiments with other advanced representative networks on five public datasets: X-ray, Lung, ISIC-2016, ISIC-2018, and capsule endoscopy and gastrointestinal segmentation. We also performed ablation experiments on the first four datasets. The experimental results outperformed the other compared networks, confirming the effectiveness of MCI-Net. This research provides a valuable reference for achieving lightweight, accurate, and high-performance medical image segmentation network models.

Pine wilt disease (PWD) is a serious threat to pine forests. Combining unmanned aerial vehicle (UAV) images and deep learning (DL) techniques to identify infected pines is the most efficient method to determine the potential spread of PWD over a large area. In particular, image segmentation using DL obtains the detailed shape and size of infected pines to assess the disease’s degree of damage. However, the performance of such segmentation models has not been thoroughly studied. We used a fixed-wing UAV to collect images from a pine forest in Laoshan, Qingdao, China, and conducted a ground survey to collect samples of infected pines and construct prior knowledge to interpret the images. Then, training and test sets were annotated on selected images, and we obtained 2352 samples of infected pines annotated over different backgrounds. Finally, high-performance DL models (e.g., fully convolutional networks for semantic segmentation, DeepLabv3+, and PSPNet) were trained and evaluated. The results demonstrated that focal loss provided a higher accuracy and a finer boundary than Dice loss, with the average intersection over union (IoU) for all models increasing from 0.656 to 0.701. From the evaluated models, DeepLLabv3+ achieved the highest IoU and an F1 score of 0.720 and 0.832, respectively. Also, an atrous spatial pyramid pooling module encoded multiscale context information, and the encoder–decoder architecture recovered location/spatial information, being the best architecture for segmenting trees infected by the PWD. Furthermore, segmentation accuracy did not improve as the depth of the backbone network increased, and neither ResNet34 nor ResNet50 was the appropriate backbone for most segmentation models.

High-order possibilistic c-means algorithms based on tensor decompositions for big data in IoT

Towards Foundation Models and Few-Shot Parameter-Efficient Fine-Tuning for Volumetric Organ Segmentation.

Current developments in machine learning, particularly related to deep learning, are proving instrumental in identification, and quantification of patterns in the medical images. The pivotal point of these advancements is the essential capability of the deep learning approaches to obtain hierarchical feature representations directly from the images, which in turn is eliminating the need for handcrafted features. Deep learning is expeditiously turning into the state-of-the-art for medical image processing and has resulted in performance improvements in diverse clinical applications. We aim to create a diagnostic method, optimized by the use of deep learning (DL) and validated by a controlled clinical trial, randomized, is a highly automated tool for diagnosing and staging precancerous and cervical cancer and thyroid cancers. We want to create a high-performance deep learning model for medical picture segmentation that is independent of the type of organs/tissues, dimensions, or image type (2D/3D) and validate it in a randomised, controlled clinical trial using U-Net-based architectures. We largely employed U-Net-based architecture analysis as a technique to identify the major features that we thought crucial in the design and optimization of the integrated DL model, based on the U-Net-based architectures that we imagined. Second, we'll use a randomised, controlled clinical trial to validate the DL model's performance. Our DL model will be a highly automated tool for diagnosing and staging precancers, cervical cancer, and thyroid cancer. The combined model we designed takes into account the key features of each of the architectures Overcomplete Convolutional Network Kite-Net (Kite-Net), Attention gate mechanism is an improvement added on convolutional network architecture for fast and precise segmentation of images (Attention U-Net), Harmony Densely Connected Network-Medical image Segmentation (HarDNet-MSEG). In this regard, we will develop a complete computer-assisted diagnosis technique that will be validated through a randomised controlled experiment. The model will be a highly automated diagnostic and staging tool for precancers, cervical cancer, and thyroid cancer. This would help specialists save time and effort when analysing medical images, aid in the development of a better therapeutic strategy, and provide a "second opinion" on computer-assisted diagnosis.