Multimodal approach to optimize biopsy decision-making for PI-RADS 3 lesions on multiparametric MRI

To develop and evaluate a multimodal approach including clinical parameters and biparametric MRI-based artificial intelligence (AI) model for determining the necessity of prostate biopsy in patients with PI-RADS 3 lesions.

This retrospective study included a prospectively recruited patient cohort with PI-RADS 3 lesions who underwent prostate MRI and MRI/US fusion-guided biopsy between April 2019 and February 2024 in a single institution. The study examined demographic data, PSA and PSA density (PSAD) levels, prostate volumes, prospective PI-RADS v2.1-compliant interpretations of a genitourinary radiologist, lesion characteristics, history of prior biopsies, and AI evaluations, focusing mainly on the detection of clinically significant prostate cancer (csPCa) (International Society of Urological Pathology grade group ≥2) on MRI/US fusion-guided biopsy. The AI model lesion segmentations were compared to manual segmentations and biopsy results. The statistical methods employed included Fisher's exact test and logistic regression.

The cohort was comprised of 248 patients with 312 PI-RADS 3 lesions in total (n = 268 non-csPCa, n = 44 csPCa). The AI model's negative predictive value (NPV) was 89.2 % for csPCa in all lesions. In patient-level analysis, the NPV was 91.2 % for patients with a highest PI-RADS score of 3. PSAD was a significant predictor of csPCa (odds ratio = 5.8, p = 0.038). Combining AI and PSAD, where AI correctly mapped a lesion or PSAD ≥0.15 ng/mL², achieved higher sensitivity (77.8 %) while maintaining a high NPV (93.1 %).

Combining AI and PSAD has the potential to enhance biopsy decision-making for PI-RADS 3 lesions by minimizing missed csPCa occurrences and reducing unnecessary biopsies.