Post

New research · Pathology
Acta radiologica (Stockholm, Sweden : 1987) · 1w
AI / informaticsActa radiologica (Stockholm, Sweden : 1987) · 2026

Artificial intelligence in prostate MRI: Comparative diagnostic performance in a high-prevalence cohort.

Nádia Gonçalves Ferreira, Owen Matthew Truscott Thomas, Kjell-Inge Gjesdal … Jonn Terje Geitung
Read paper
PathologyAI / informatics

Artificial intelligence missed more clinically significant prostate cancers on MRI than radiologists

Artificial intelligence in prostate MRI: Comparative diagnostic performance in a high-prevalence cohort.

Nádia Gonçalves Ferreira … Jonn Terje Geitung
Acta radiologica (Stockholm, Sweden : 1987) · 2026
Background

Artificial intelligence (artificial intelligence) is increasingly used in prostate cancer diagnostic workflows but remains insufficiently validated in high-prevalence cohorts often encountered in academic referral centers.

Purpose

To compare the diagnostic performance of licensed artificial intelligence software with routine radiologist readings of prostate MRI, using histopathology as the reference standard.

Methods

Material and Methods In this retrospective study, 1000 patients underwent prostate MRI for suspected prostate cancer (between May 2020 and December 2024), followed by transperineal biopsy; 391 subsequently underwent radical prostatectomy.

n = 1000 patients
Results

Artificial intelligence was more likely than radiologists to miss serious prostate cancers on MRI

96%
Sensitivity
20.9%
Specificity
More results

Artificial intelligence assigned more cases to PI-RADS 1-2 and fewer to PI-RADS 3 (κ = 0.388; P <.001).

At the PI-RADS ≥3 threshold, radiologists showed sensitivity 96.0%, specificity 20.9%, accuracy 85.9%, positive predictive value 88.7%, and negative predictive value 45.0%, compared with 91.3%, 37.2%, 84.0%, 90.3%, and 40.0% for artificial intelligence ( P <.001).

More results

ConclusionAI showed diagnostic performance comparable to that of radiologists for csPCa detection, with no significant difference in AUC.

“
Conclusion

Artificial intelligence reduced indeterminate PI-RADS 3 scores but missed more significant cancers.

Read paper
0 comments

No comments yet. Be the first.

Related papers

LatestFoundational
Cohort Study
15.6%
about 1 in 6 with the deepest submucosal invasion had cancer in nearby lymph nodes
Cohort Study
r=-0.23
Weak negative correlation: lower eyelid-lifting strength, more fat within the muscle tissue
Cohort Study
AUC 0.81
Good ability to separate cancer-containing from cancer-free underarm lymph nodes
Cohort Study
82%
Share of disagreeing cases showing uneven HER2 staining within one tumor block
Animal / Preclinical
< -7.0 kcal/mol
a lower value means stronger attachment to AMPK/mTOR proteins
Animal / Preclinical
6.3 × 4.1 × 3.6 mm
the dimensions of the nasal stone, which was large for the dog's nasal passage
Cohort Study
52%
More than half of patients had sudden, severe damage to their liver cells
Cohort Study
50.058-fold
having EGFR makes the risk of aggressive laryngeal cancer substantially higher