Emerging evidence
More
AI flagged all 5 cancer-positive prostate resection specimens in an 11-specimen study
Study title: Clinical Utility of Tumor-Detect Artificial Intelligence Algorithm for Detection of Prostate Cancer on Transurethral Resection Specimen
- Evidence type
- Benchmark or lab evaluation
- Sample
- n = 81
- Sources
- 1 sourcePeer-reviewed
- Published
- 2 min read
- Truvace record
- TRV-2026-2909cite or check it
Study title: Clinical Utility of Tumor-Detect Artificial Intelligence Algorithm for Detection of Prostate Cancer on Transurethral Resection Specimen
A study in the International Journal of Surgical Pathology tested PathAI's Tumor-Detect algorithm on 81 whole-slide images from 11 transurethral resection of prostate (TURP) specimens. The algorithm flagged all five specimens with incidental prostate adenocarcinoma, a specimen-level sensitivity of 100%.
Read the study abstract
Artificial intelligence (AI)-based tumor detection algorithms are increasingly being integrated into digital pathology workflows. This study evaluated the performance of the Tumor-Detect AI algorithm for identifying prostate cancer in transurethral resection of prostate (TURP) specimens, which are diagnostically challenging due to tissue fragmentation and small incidental tumor foci. Eighty-one whole-slide images from 11 TURP specimens, including six specimens diagnosed as benign prostatic hyperplasia and five s…

The gain
In a study of 81 slides from 11 prostate resection specimens, PathAI's Tumor-Detect algorithm identified all five cancer-positive specimens (100% specimen-level sensitivity), with slide-level sensitivity of 88% and specificity of 86% against a pathologist's ground truth.
In brief
At the level of individual slides, sensitivity was 88% and specificity 86%, measured against a genitourinary pathologist's reading. TURP specimens are hard to read because the tissue is fragmented and tumor foci can be small and incidental, so a tool that flags suspicious slides could help pathologists avoid missing cancers.
The sample was very small, with only six benign and five cancer specimens. The false positives came from complex glandular architecture, clear cell change and high-grade prostatic intraepithelial neoplasia, which shows where the tool still mistakes benign or precancerous features for cancer.
Main points
- The study analysed 81 whole-slide images from 11 TURP specimens: six diagnosed as benign prostatic hyperplasia and five with incidental prostate adenocarcinoma.
- The Tumor-Detect algorithm from PathAI of Boston flags slides with suspected malignancy and generates heatmaps highlighting tumor regions.
- All five malignant specimens were correctly identified, giving 100% specimen-level sensitivity.
- Slide-level sensitivity was 88% and specificity 86% compared with a genitourinary pathologist's ground truth.
- False positives were mainly linked to complex glandular architecture, clear cell change or high-grade prostatic intraepithelial neoplasia.
The full story
The researchers chose TURP specimens because they are diagnostically challenging: the resected tissue is fragmented, and any cancer present is often small and incidental. The study set was 11 specimens, split between six benign prostatic hyperplasia cases and five with incidental adenocarcinoma.
Each of the 81 whole-slide images was run through Tumor-Detect, which flags slides it suspects contain malignancy and produces heatmaps over tumor regions. A genitourinary pathologist established the ground truth for comparison.
Every cancer-positive specimen was caught, and slide-level sensitivity and specificity were 88% and 86%. The authors say the tool may improve efficiency and reduce missed diagnoses in routine pathology, though they report false positives from benign look-alike patterns and precancerous change.
Sources
1 on file- International Journal of Surgical Pathologydoi:10.1177/10668969261493447
Talk it over
Tumor-Detect flagged all five cancer-positive specimens but also produced false positives from clear cell change and high-grade prostatic intraepithelial neoplasia. How would you weigh catching every cancer against those false alarms?
AI flagged all 5 cancer-positive prostate resection specimens in an 11-specimen studyhttps://truvace.com/item/clinical-utility-of-tumor-detect-artificial-intelligence-algorithm-for-detection
Tumor-Detect flagged all five cancer-positive specimens but also produced false positives from clear cell change and high-grade prostatic intraepithelial neoplasia. How would you weigh catching every cancer against those false alarms?
Answer in the threadThe study used 81 slides from 11 specimens, only six benign and five with cancer, judged against one genitourinary pathologist. How much can 100% specimen-level sensitivity tell us with a sample that size?
Answer in the threadIf you were a pathologist reading fragmented TURP tissue with Tumor-Detect heatmaps beside you, how would you decide when to trust a flagged slide and when to overrule it?
Answer in the thread
If you have more time
What kind of larger study, with how many specimens and how many pathologists, would settle whether Tumor-Detect reduces missed cancers in routine work?
Answer in the threadLook up what happens to patients when a small incidental cancer in a TURP specimen is missed. How would that change the weight you give to false positives?
Answer in the thread
ace
What readers say