USMF logo

Institutional Repository in Medical Sciences
of Nicolae Testemitanu State University of Medicine and Pharmacy
of the Republic of Moldova
(IRMS – Nicolae Testemitanu SUMPh)

Biblioteca Stiintifica Medicala
DSpace

University homepage  |  Library homepage

 
Please use this identifier to cite or link to this item: http://hdl.handle.net/20.500.12710/34258
Title: ARTIFICIAL INTELLIGENCE IN EQUIVOCAL PI-RADS LESION ASSESSMENT: SYSTEMATIC REVIEW AND META ANALYSIS
Authors: Borcoman, Marian
Leşan, Octavian
Cazacu, Oleg
Scutari, Adrian
Keywords: prostate cancer;artificial intelligence;PI-RADS;mpMRI;equivocal lesions
Issue Date: 2026
Publisher: Centrul Editorial-Poligrafic Medicina
Citation: Borcoman, Marian; Leşan, Octavian; Cazacu, Oleg; Scutari, Adrian. ARTIFICIAL INTELLIGENCE IN EQUIVOCAL PI-RADS LESION ASSESSMENT: SYSTEMATIC REVIEW AND META ANALYSIS. In: MedEspera 2026, p. 442. ISBN 978-9975-82-486-6.
Abstract: Introduction. PI-RADS 3 and borderline PI-RADS 4 lesions account for 30-40% of mpMRI findings, creating substantial diagnostic uncertainty in urological practice. Inter-observer variability remains high (kappa 0.45-0.60), generating uncertainty and leading to biopsy recommendations in 60-80% of cases, with benign results in 40-60% and csPCa miss rates up to 25% in negative biopsies. Aim of study. To systematically review the latest studies that are evaluating AI models for reclassification of equivocal PI-RADS lesions, quantifying diagnostic accuracy, negative predictive value, and reduction in unnecessary biopsies. Methods and materials. A narrative review and meta-analysis were conducted. Analysis is based on 10 studies carried out between 2024-2026. Deep learning algorithms (CNN/transformer) analyzed T2-weighted, diffusion-weighted and dynamic contrast-enhanced MRI sequences. Primary outcomes included reclassification rates, negative predictive value for biopsy avoidance, diagnostic sensitivity/specificity, and biopsy reduction. Results. In 65% of PI-RADS 3 lesions have been reclassified by AI algorithms: 61% have been downgraded to PI-RADS 2 or less (pooled NPV 95.8%, CI 95%) avoiding biopsy; on the other hand 39% upgraded to PI-RADS 4 or more, requiring biopsy. PI-RADS 4 lesions saw 27% downgrading (PPV 92.1%). Diagnostic performance reached sensitivity 94.2% (IQR 91- 97%) and specificity 86.7% (IQR 84-90%), surpassing radiologist baseline readings (78%/79%). Biopsy reduction averaged 39% (from 78% to 46% of equivocal cases; OR=0.61, 95% CI 0.52- 0.72), while csPCa detection remained stable at 51.2%. Inter-observer agreement improved from k=0.59 to 0.87. Processing averaged 28 seconds per case. Conclusions. Equivocal PI-RADS lesions remain a challenge for selecting the appropriate diagnosis and treatment strategy, but AI reassessment demonstrates superior precision and clinical efficiency, enabling avoidance of 39% unnecessary biopsies without compromising csPCa detection. Prospective multicenter validation will facilitate integration into standard urological protocols.
metadata.dc.relation.ispartof: MedEspera 2026. The 11th International Medical Congress for Students and Young Doctors, 6-8 May 2026
URI: https://medespera.md/documents/abstract-book-4.pdf
https://repository.usmf.md/handle/20.500.12710/34258
ISBN: 978-9975-82-486-6
Appears in Collections:MedEspera 2026

Files in This Item:
File Description SizeFormat 
M_442.pdf466.41 kBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

 

Valid XHTML 1.0! DSpace Software Copyright © 2002-2013  Duraspace - Feedback