Data-Efficient Prostate Micro-Ultrasound Image Analysis using a Modality-specific Foundation Model
- 1UCLA
Abstract
Background and ObjectiveMicro-ultrasound (MicroUS) is an emerging high-resolution imaging modality for prostate cancer diagnosis and real-time biopsy guidance, but its automated analysis is limited by labor-intensive expert annotation. We aimed to develop and evaluate a foundation model for data-efficient MicroUS image analysis.
MethodsWe developed {micro}FM, a MicroUS-specific foundation model pretrained using DINOv2-style self-supervised learning on 1,736,391 unlabeled images from 780 patients. The model was fine-tuned and evaluated in three downstream cohorts for prostate capsule segmentation (public, 75 patients), urethra segmentation (institutional, 105 patients), and prostate cancer lesion localization (institutional, 190 patients). Under limited annotation, {micro}FM was compared with randomly initialized models and foundation models pretrained on natural, medical, and ultrasound images. Evaluation used the Dice scores and 95th-percentile Hausdorff distance (HD95) for segmentation and free-response receiver operating characteristic analysis for lesion localization.
Key Findings and Limitations{micro}FM improved performance across all three tasks, with the largest gains under limited annotation. With 10% of training scans, {micro}FM achieved Dice scores of 93.8% for prostate capsule segmentation and 58.4% for urethra segmentation, versus 87.9% and 47.0% for the strongest comparators. Corresponding HD95 values were 2.0 and 4.3 mm. For lesion localization, {micro}FM outperformed other models with approximately 0.75 sensitivity at two false-positive detections per 3D volume. Limitations include retrospective single-institution pretraining and lack of explicit 3D modeling.
Conclusions and Clinical ImplicationsMicroUS-specific self-supervised pretraining improved data-efficient image analysis across prostate and urethra segmentation, and lesion localization. This approach may reduce annotation requirements for developing automated tools in MicroUS image analysis.
Advancing PracticeO_ST_ABSWhat does the study add?C_ST_ABSThis study evaluates a deep learning model pretrained on a large-scale dataset of unlabeled prostate MicroUS images. The pretrained model was adapted to three clinically relevant image-analysis tasks: prostate capsule segmentation, urethra segmentation, and prostate cancer lesion localization. Substantial performance gains over other models were observed despite limited training data, suggesting that MicroUS-specific self-supervised pretraining may reduce expert annotation requirements for developing automated image-analysis tools.
Patient SummaryWe developed a foundation model that learns from unlabeled prostate MicroUS images. It performed better with limited labeled data but requires independent validation before clinical use.
Related
- Generative AI in Radiotherapy Imaging: From Cross-Modality Synthesis to Foundation Model-Based Segmentation.
- Beyond a diagnostic modality: the pivotal role of ultrasound biomicroscopy in ophthalmic care pathways.
- Vaginal Laser Therapy for Genitourinary Syndrome of Menopause in Breast Cancer Survivors: A Modality-Specific Systematic Review of Efficacy, Durability, and Safety
- Artificial intelligence-based prostate cancer risk stratification via large-scale ultrasound images paired with histopathology
- Data-efficient prostate cancer classification from T2-weighted MRI with mechanistic auditing and a dual-agent quality-control workflow
BibTeX
@misc{pang2026dataefficient,
title = {{Data-Efficient Prostate Micro-Ultrasound Image Analysis using a Modality-specific Foundation Model}},
author = {K. Pang and J. Fan and A. L. Y. Hung and Q. Miao and C. Oh and Q. He and B. Wodlinger and A. Kinnaird and W. Brisbane and K. Sung},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.29.26364330},
url = {https://www.medrxiv.org/content/10.64898/2026.09.29.26364330v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Data-Efficient Prostate Micro-Ultrasound Image Analysis using a Modality-specific Foundation Model
AU - K. Pang
AU - J. Fan
AU - A. L. Y. Hung
AU - Q. Miao
AU - C. Oh
AU - Q. He
AU - B. Wodlinger
AU - A. Kinnaird
AU - W. Brisbane
AU - K. Sung
PY - 2026
DA - 2026/09/29
PB - medRxiv
DO - 10.64898/2026.09.29.26364330
UR - https://www.medrxiv.org/content/10.64898/2026.09.29.26364330v1
AB - Background and ObjectiveMicro-ultrasound (MicroUS) is an emerging high-resolution imaging modality for prostate cancer diagnosis and real-time biopsy guidance, but its automated analysis is limited by labor-intensive expert annotation. We aimed to develop and evaluate a foundation model for data-efficient MicroUS image analysis. MethodsWe developed {micro}FM, a MicroUS-specific foundation model pretrained using DINOv2-style self-supervised learning on 1,736,391 unlabeled images from 780 patients. The model was fine-tuned and evaluated in three downstream cohorts for prostate capsule segmentation (public, 75 patients), urethra segmentation (institutional, 105 patients), and prostate cancer lesion localization (institutional, 190 patients). Under limited annotation, {micro}FM was compared with randomly initialized models and foundation models pretrained on natural, medical, and ultrasound images. Evaluation used the Dice scores and 95th-percentile Hausdorff distance (HD95) for segmentation and free-response receiver operating characteristic analysis for lesion localization. Key Findings and Limitations{micro}FM improved performance across all three tasks, with the largest gains under limited annotation. With 10% of training scans, {micro}FM achieved Dice scores of 93.8% for prostate capsule segmentation and 58.4% for urethra segmentation, versus 87.9% and 47.0% for the strongest comparators. Corresponding HD95 values were 2.0 and 4.3 mm. For lesion localization, {micro}FM outperformed other models with approximately 0.75 sensitivity at two false-positive detections per 3D volume. Limitations include retrospective single-institution pretraining and lack of explicit 3D modeling. Conclusions and Clinical ImplicationsMicroUS-specific self-supervised pretraining improved data-efficient image analysis across prostate and urethra segmentation, and lesion localization. This approach may reduce annotation requirements for developing automated tools in MicroUS image analysis. Advancing PracticeO_ST_ABSWhat does the study add?C_ST_ABSThis study evaluates a deep learning model pretrained on a large-scale dataset of unlabeled prostate MicroUS images. The pretrained model was adapted to three clinically relevant image-analysis tasks: prostate capsule segmentation, urethra segmentation, and prostate cancer lesion localization. Substantial performance gains over other models were observed despite limited training data, suggesting that MicroUS-specific self-supervised pretraining may reduce expert annotation requirements for developing automated image-analysis tools. Patient SummaryWe developed a foundation model that learns from unlabeled prostate MicroUS images. It performed better with limited labeled data but requires independent validation before clinical use.
ER -