In our recently published study in Annals of Nuclear Medicine, we assessed the diagnostic accuracy and efficiency of aPROMISE compared to manual PROMISE V2-based reads in Japanese patients with metastatic prostate cancer (mPCa), utilizing both ^68Ga-PSMA-11 and ^18F-PSMA-1007 PET/CT.
We analyzed 21 PET/CT scans (12 using ^68Ga, 9 using ^18F), interpreted by a single experienced nuclear radiologist. Each scan was reviewed first manually, and then with aPROMISE assistance. Diagnostic times and miTNM concordance rates were compared.
Strengths of the Study
- First-of-its-kind validation of aPROMISE using both ^68Ga and ^18F tracers in a Japanese population:
- Most prior studies evaluating aPROMISE have focused solely on ^18F-based tracers in Western cohorts. This study uniquely assesses its diagnostic utility in ^68Ga-PSMA-11 and ^18F-PSMA-1007 PET/CT among Japanese patients with mPCa, addressing a critical evidence gap in AI application across diverse ethnic populations and tracer types.
- Objective measurement of diagnostic efficiency:
- The study clearly demonstrates that aPROMISE significantly reduces diagnostic time (by over 60%) compared to manual reads, particularly with ^68Ga-PSMA PET/CT, indicating potential to streamline clinical workflows and reduce physician burden.
- Detailed comparative analysis of miTNM staging performance:
- By evaluating sensitivity and specificity across all miTNM components, the study offers a granular view of where aPROMISE excels (miT, M1a, M1b) and where limitations remain (M1c, especially visceral lesions).
- Integration of AI-derived quantitative metrics:
- aPROMISE provides lesion counts, volumes, and SUV statistics—parameters increasingly important for treatment stratification, especially in the context of PSMA-targeted radioligand therapy (RLT).
- Small sample size and single-center, retrospective design:
- With only 21 patients and a single nuclear radiologist performing all reads, generalizability is limited. Inter-reader variability could not be assessed.
- No histopathological confirmation of lesions:
- The study defines manual reads as the “gold standard,” but without pathological validation, there is a possibility of under- or over-estimation of true metastatic burden.
- Challenges in visceral lesion detection by aPROMISE:
- AI misclassified or missed several adrenal and liver metastases, likely due to physiological uptake masking true lesions—a known limitation in PSMA PET/CT interpretation that AI has yet to overcome.
- Tracer-specific differences not statistically powered:
- Although the study includes both ^68Ga and ^18F tracers, comparisons between them remain descriptive due to insufficient sample size per group.
This study supports the clinical utility of aPROMISE as an efficient and accurate AI-assisted platform for miTNM staging in Japanese patients with mPCa. While its performance in detecting bone and nodal metastases is excellent—achieving 100% sensitivity and specificity for M1b lesions—its limitations in evaluating visceral disease underscore the need for continued human oversight.
Moving forward, larger, multicenter prospective studies are essential to validate the diagnostic robustness of aPROMISE across different radioligands and patient populations. Integration of AI-derived quantitative metrics with clinical outcomes (e.g., response to PSMA-RLT) may further enhance personalized care in advanced prostate cancer. Additionally, refining AI algorithms to better distinguish physiological from pathological uptake—especially in the liver and adrenal glands—remains a key development goal.
Written by: Yuki Enei, MD, PhD, Department of Urology, Jikei University School of Medicine, Minato, Japan
Read the Abstract