BJUI compass · 2026
MUSeg-PSV: A Real-Time Deep-Learning Segmentation of the Prostate Gland and Seminal Vesicles on 29-MHz Micro-Ultrasound.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Cella Ludovica, Paciotti Marco, Cavadini Giacomo, Di Stefano Luca, Avolio Pier Paolo, Fasulo Vittorio, Piccolini Andrea, Saitta Cesare, Beatrici Edoardo, Kocielnik Rafal, Hung Andrew J, Saita Alberto, Casale Paolo, Lazzeri Massimo, Lughezzani Giovanni, Buffi Nicolò Maria
Аннотация
OBJECTIVES: This study aimed to develop and evaluate MUSeg-PSV, a real-time deep-learning framework for segmentation of the prostate gland and seminal vesicles on 29-MHz micro-ultrasound (MicroUS), intended to support anatomical orientation during MicroUS-guided prostate biopsy and local staging of prostate cancer.
PATIENTS AND METHODS: MicroUS data from 14 patients (21 full-length videos; 2588 annotated frames) were used to train and test MUSeg-PSV, a lightweight single-stage convolutional model based on YOLOv11s-seg. Two training strategies were compared: a standard configuration with basic augmentations and an augmentation-rich configuration incorporating advanced spatial and photometric transformations. Model performance was assessed on a held-out test set using mean average precision (mAP), dice similarity coefficient (DSC), expected calibration error (ECE) and Brier score. Real-time feasibility was evaluated through inference latency and qualitative assessment of temporal consistency on unseen videos.
RESULTS: The augmentation-rich model outperformed the standard configuration, yielding a 21% relative increase in mAP and a 4.6-fold improvement in seminal vesicle AP (0.183 → 0.842). Prostate DSC increased from 0.675 to 0.770 (p = 4.4 × 10-5), while seminal vesicle DSC remained robust (0.892 vs. 0.876), reflecting a controlled trade-off with substantially improved SV detection (AP: 0.183 → 0.842). Calibration improved markedly, with ECE decreasing from 0.558 to 0.156 and Brier score from 1.000 to 0.196 (p < 0.01). Qualitative evaluation confirmed smooth, temporally coherent overlays and consistent delineation of challenging anatomical regions. End-to-end latency (28-35 ms/frame; ~35 fps) demonstrated compatibility with real-time clinical deployment.
CONCLUSION: MUSeg-PSV enables reliable, real-time multiclass segmentation of high-frequency MicroUS, providing stable delineation of the prostate and seminal vesicles at clinically viable frame rates. These results support the potential of AI-assisted MicroUS to enhance operator orientation during MicroUS-guided prostate biopsy and local staging, and to promote standardised identification of seminal vesicles within prostate cancer diagnostic and staging workflows.
PATIENTS AND METHODS: MicroUS data from 14 patients (21 full-length videos; 2588 annotated frames) were used to train and test MUSeg-PSV, a lightweight single-stage convolutional model based on YOLOv11s-seg. Two training strategies were compared: a standard configuration with basic augmentations and an augmentation-rich configuration incorporating advanced spatial and photometric transformations. Model performance was assessed on a held-out test set using mean average precision (mAP), dice similarity coefficient (DSC), expected calibration error (ECE) and Brier score. Real-time feasibility was evaluated through inference latency and qualitative assessment of temporal consistency on unseen videos.
RESULTS: The augmentation-rich model outperformed the standard configuration, yielding a 21% relative increase in mAP and a 4.6-fold improvement in seminal vesicle AP (0.183 → 0.842). Prostate DSC increased from 0.675 to 0.770 (p = 4.4 × 10-5), while seminal vesicle DSC remained robust (0.892 vs. 0.876), reflecting a controlled trade-off with substantially improved SV detection (AP: 0.183 → 0.842). Calibration improved markedly, with ECE decreasing from 0.558 to 0.156 and Brier score from 1.000 to 0.196 (p < 0.01). Qualitative evaluation confirmed smooth, temporally coherent overlays and consistent delineation of challenging anatomical regions. End-to-end latency (28-35 ms/frame; ~35 fps) demonstrated compatibility with real-time clinical deployment.
CONCLUSION: MUSeg-PSV enables reliable, real-time multiclass segmentation of high-frequency MicroUS, providing stable delineation of the prostate and seminal vesicles at clinically viable frame rates. These results support the potential of AI-assisted MicroUS to enhance operator orientation during MicroUS-guided prostate biopsy and local staging, and to promote standardised identification of seminal vesicles within prostate cancer diagnostic and staging workflows.