ProFound: A moderate-sized vision foundation model for multi-task prostate imaging

Yipei Wang; Yinsong Xu; Weixi Yi; Shaheer Ullah Saeed; Natasha Thorley; Alexander Ng; Yukun Zhou; Wen Yan

cs.CVMar 4, 2026

ProFound: A moderate-sized vision foundation model for multi-task prostate imaging

Authors: Yipei Wang, Yinsong Xu, Weixi Yi, Shaheer Ullah Saeed, Natasha Thorley, Alexander Ng, Yukun Zhou, Wen Yan, +6 more

Organizations: UCL Hawkes Institute, University College London, London, UK · Department of Medical Physics and Biomedical Engineering, University College London, London, UK · School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China · Centre for Bioengineering, Queen Mary University of London, London, UK · School of Engineering and Materials Science, Queen Mary University of London, London, UK · Digital Environment Research Institute, Queen Mary University of London, London, UK · Centre for Medical Imaging, University College London, London, UK · Department of Radiology, University College London Hospital NHS Foundation Trust, London, UK · Centre for Urology Imaging, Prostate, AI and Surgical Studies (COMPASS) Research Group, Division of Surgery and Interventional Science, University College London, London, UK · Division of Surgery and Interventional Science, University College London, London, UK · Institute of Ophthalmology, University College London, London, UK · Department of Urology, University College London Hospital, London, UK · Department of Urology, Comprehensive Cancer Center, Medical University of Vienna, Vienna, Austria · Department of Computer Science, University College London, London, UK

Abstract

Many diagnostic and therapeutic clinical tasks for prostate cancer increasingly rely on multi-parametric MRI. Automating these tasks is challenging because they necessitate expert interpretations, which are difficult to scale to capitalise on modern deep learning. Although modern automated systems achieve expert-level performance in isolated tasks, their general clinical utility remains limited by the requirement of large task-specific labelled datasets. In this paper, we present ProFound, a domain-specialised vision foundation model for volumetric prostate mpMRI. ProFound is pre-trained using several variants of self-supervised approaches on a diverse, multi-institutional collection of 5,000 patients, with a total of over 22,000 unique 3D MRI volumes (over 1,800,000 2D image slices). We conducted a systematic evaluation of ProFound across a broad spectrum of $11$ downstream clinical tasks on over 3,000 independent patients, including prostate cancer detection, Gleason grading, lesion localisation, gland volume estimation, zonal and surrounding structure segmentation. Experimental results demonstrate that finetuned ProFound consistently outperforms or remains competitive with state-of-the-art specialised models and existing medical vision foundation models trained/finetuned on the same data.

Figures & tables

Figure 1: Overview of the proposed prostate mpMRI foundation model.

Task name	Description	Dataset	Subtask	Cat.
PIRADS-Cls	Automating radiologist task by predicting	ReIMAGINE	PIRADS $\leq 3$ , 3, 4, 5	mul-Cls
	PIRADS score	-Risk	PIRADS $<3$ vs. $\geq 3$	bin-Cls
			PIRADS $\leq 3$ vs. $>3$	bin-Cls
Grading-Cls	Predicting Gleason Grade Group on patients	PROMIS	GG $\leq 6$ ,3+4,4+3,8,>8	mul-Cls
	with MR-positive lesions		GG $\leq 3+4$ vs. $\leq 6$	bin-Cls
			GG $\leq 4+3$ vs. $\leq 3+4$	bin-Cls

Table 1: Description of ProFound downstream tasks and extended analysis. Cat. = Category, Cls = classification, Seg = segmentation, Reg = regression, Loc = localisation, mul = multi-class, bin = binary, GG = Gleason group, FM = Foundation model

Figure 2: Performance on representative downstream tasks. ProFound achieves the highest Dice scores on both lesion segmentation tasks, while remaining comparable to SOTA specialised models on anatomy segmentation and prostate volume regression. ProFound shows a clear advantage in limited-label settings, demonstrating strong data efficiency. We also include a challenging task where ProFound provides more nuanced gains.

	PIRADS-Cls				Grading-Cls				Cls	PCa-Loc			Loc
	QWK	AUC	Spec	Sens	QWK	AUC	Spec	Sens	Params	AUC	Spec	Sens	Params
Baseline	0.207	73.2	50.0	49.1	-0.002	59.1	35.4	22.4	33.3	61.3	41.4	25.0	4.8
ViT	0.224	72.1	36.1	56.1	-0.035	47.4	20.8	13.2	123.0	60.5	41.3	26.2	136.3
ConvNeXt	0.010	59.7	38.9	22.2	0.087	59.6	39.6	21.9	29.5	62.4	43.2	30.0	39.5
RadFM	-0.002	73.4	47.2	62.6	0.013	55.7	25.0	18.9	66.4	60.3	38.5	22.7	79.6
RadDiag	0.076	73.3	55.6	54.4	0.147	72.0	47.9	50.4	33.3	63.5	42.7	27.0	119.4

Table 2: Performance comparison for PIRADS score classification, Gleason grade group classification, and localised cancer detection task. Spec = Specificity at 80% sensitivity; Sens = Sensitivity at 80% specificity, + denotes ProFound with frozen encoder (linear probing). The baseline model is ResNet18 for PIRADS-Cls and Grading-Cls tasks, and UNet for the PCa-Loc localisation task. The unit for the number of updated parameters is million.

Explore similar work

Zero-Shot Adaptation of Medical Vision Foundation Models for High-Frequency Micro-Ultrasound Prostate Segmentation
Aug 14, 2026Ayusha Abbas, Saram Abbas, Kabita AdhikariUltrasound Image Segmentation Medical Image Segmentation
Newcastle University Newcastle Upon Tyne, UK
Automated Prostate Gland Segmentation in MRI Using nnU-Net
Apr 2, 2026Pablo Rodriguez-Belenguer, Gloria Ribas, Javier Aquerreta Escribano +3Medical Image Segmentation Magnetic Resonance Imaging
Biomedical Imaging Research Group (GIBI230), Instituto de Investigación Sanitaria La Fe, Valencia, Spain
Learning Where to Look: A Reinforcement Learning Framework for Robust Micro-Ultrasound Prostate Cancer Detection
Jun 29, 2026Mohammad Mahdi Abootorabi, Sina Namazi, Armin Saadat +7Medical Imaging Reinforcement Learning
The University of British Columbia, Vancouver, BC, Canada · Vector Institute, Toronto, ON, Canada · Queen’s University, Kingston, ON, Canada +1

Aug 14, 2026J/K move · Enter open · S save

Zero-Shot Adaptation of Medical Vision Foundation Models for High-Frequency Micro-Ultrasound Prostate Segmentation

Ayusha Abbas, Saram Abbas, Kabita Adhikari

Newcastle University Newcastle Upon Tyne, UK

Ultrasound Image Segmentation Medical Image Segmentation Zero-Shot Medical Image Segmentation Boundary-Aware Segmentation Medical Imaging Foundation Models

Open paper AskPDF arXiv

Prostate cancer claims a life every 80 seconds. Early detection is needed to prevent disease progression, and both PSA density calculation and biopsy decisions rely on knowing the exact boundary of the gland. Conventional ultrasound at 6-12 MHz blurs this boundary, missing one in three high-risk cancers. Micro-ultrasound (29 MHz) improves resolution threefold but introduces dense acoustic speckle that obscures the outer wall; given the same image, two clinicians draw outlines differing by over 10% in area. Supervised methods are costly and generalise poorly across scanners. Can a foundation model segment the prostate with no training data? We present the first zero-shot pipeline for this modality: MedSAM, pre-trained on over 1.5 million medical images, localises the prostate; we then apply CLAHE to sharpen the outer wall, binary dilation to recover missed pixels, and Fourier smoothing (4 modes, s=1.05) to refine the boundary. MedSAM requires a spatial prompt, so we evaluate bounding-box and point-click strategies across 75 patients of the Micro-Ultrasound Prostate Segmentation dataset (2,621 slices). On the 20-patient held-out test set, the pipeline reduces mean boundary-distance error by 45% (Dice 0.749+/-0.043 to 0.865+/-0.029; HD95 217.2+/-36.9 to 120.1+/-26.1 px), reaching Dice 0.859 across the cohort. Its mean overlap shows no significant difference from the three non-expert rater groups (p>0.19), while segmenting 38-52% more consistently (lower inter-patient standard deviation). Point-click prompts fail regardless of placement (best Dice=0.350), because speckle gives no stable local contrast. Only an approximate bounding box is required, so any clinic can deploy it without data collection, annotation, or retraining.

Abstract

Explore similar work

Zero-Shot Adaptation of Medical Vision Foundation Models for High-Frequency Micro-Ultrasound Prostate Segmentation

Automated Prostate Gland Segmentation in MRI Using nnU-Net

Learning Where to Look: A Reinforcement Learning Framework for Robust Micro-Ultrasound Prostate Cancer Detection