# AI‐enhanced micro‐ultrasound improves detection of clinically significant prostate cancer at biopsy

## TL;DR

AI-enhanced micro-ultrasound improves the detection of significant prostate cancer during biopsies, reducing false positives while maintaining high sensitivity.

## Contribution

A novel AI model using micro-ultrasound data improves prostate cancer detection accuracy compared to traditional clinical models.

## Key findings

- The AI–micro-US model achieved an AUC of 0.871 for detecting clinically significant prostate cancer.
- At a fixed threshold, the model had 92.5% sensitivity and 68.1% specificity, outperforming traditional clinical models.
- AI-enhanced micro-US shows potential as a point-of-care alternative to MRI for prostate cancer screening.

## Abstract

This study aimed to evaluate the diagnostic accuracy of artificial intelligence (AI)–enhanced micro‐ultrasound (micro‐US) for detecting clinically significant prostate cancer (csPCa) in men referred for prostate biopsy.

We retrospectively analysed 145 men undergoing micro‐US‐guided biopsy (79 with csPCa, 66 without). Deep features were extracted from 2D micro‐US slices using a self‐supervised convolutional autoencoder and classified with a random forest model under fivefold cross‐validation. Patients were considered csPCa‐positive if ≥8 consecutive slices were predicted positive. Diagnostic performance was assessed against biopsy pathology using receiver operating characteristic (ROC) analysis.

The AI–micro‐US model achieved an area under the ROC curve (AUC) of 0.871. At a fixed threshold, sensitivity was 92.5% and specificity 68.1%, outperforming a clinical model based on prostate‐specific antigen (PSA), digital rectal examination (DRE), age, and prostate volume (AUC 0.753; sensitivity 96.2%, specificity 27.3%).

AI‐enhanced micro‐US reduces false positives from conventional screening tools while preserving high sensitivity. It shows promise as a point‐of‐care alternative to MRI, integrating risk stratification and biopsy guidance into a single platform.