ProstNFound+: A Prospective Study using Medical Foundation Models for Prostate Cancer Detection

TL;DR

Prostate cancer detection from micro-ultrasound benefits from robust, scalable AI that can operate with limited expert input. This work introduces ProstNFound+, a medical foundation-model–based detector that uses adapter tuning and a PCa-aware prompt encoder to output a cancer heatmap and a csPCa risk score. In a prospective validation at a new clinical center five years after training, ProstNFound+ generalizes without performance loss and shows competitive discrimination against PRI-MUS and PI-RADS, with interpretable heatmaps reflecting biopsy-confirmed lesions. The study demonstrates the potential for clinically deployable, interpretable, and scalable AI tools to assist μUS-guided prostate cancer diagnosis, potentially reducing reliance on operator expertise and expanding access in low-resource settings.

Abstract

Purpose: Medical foundation models (FMs) offer a path to build high-performance diagnostic systems. However, their application to prostate cancer (PCa) detection from micro-ultrasound (μUS) remains untested in clinical settings. We present ProstNFound+, an adaptation of FMs for PCa detection from μUS, along with its first prospective validation. Methods: ProstNFound+ incorporates a medical FM, adapter tuning, and a custom prompt encoder that embeds PCa-specific clinical biomarkers. The model generates a cancer heatmap and a risk score for clinically significant PCa. Following training on multi-center retrospective data, the model is prospectively evaluated on data acquired five years later from a new clinical site. Model predictions are benchmarked against standard clinical scoring protocols (PRI-MUS and PI-RADS). Results: ProstNFound+ shows strong generalization to the prospective data, with no performance degradation compared to retrospective evaluation. It aligns closely with clinical scores and produces interpretable heatmaps consistent with biopsy-confirmed lesions. Conclusion: The results highlight its potential for clinical deployment, offering a scalable and interpretable alternative to expert-driven protocols.

Figures (4)

• Figure 1: ProstNFound+ integrates a B-mode image encoder with conditional prompting using clinical metadata. The resulting embeddings are used by the mask decoder to generate a heatmap of cancer likelihood, and by the class decoder to output an image-level score representing the likelihood of clinically significant prostate cancer (csPCa).
• Figure 2: (A) Ablation study results. (B) csPCa detection performance (left) and average heatmap activation (right) by true involvement of cancer in samples.
• Figure 3: Example heatmaps generated by ProstNFound+. Higher PRI-MUS and model risk scores indicate higher suspicion of cancer, and red activations indicate suspicious lesions. Suspicion of cancer typically coincides with true cancer confirmed by biopsy.
• Figure 4: Left: Risk scores and results across biopsies for subjects in the prospective test set. Row groups distributed vertically represent subjects, and horizontal position within row group represents biopsy number. Subjects are grouped into columns by subject-level diagnosis. Right: breakdown of subject-level diagnoses (no cancer, isPCacsPCa) by highest PRI-MUS score and highest model risk score for that patient.