AI Boosts Prostate Cancer Detection on MRI Scans

In a groundbreaking diagnostic study, artificial intelligence (AI) assistance has proven to significantly enhance the detection of prostate cancer on MRI scans. The study, conducted by an international team of researchers, found that AI support increased diagnostic accuracy by 3.3% compared to unassisted readings. This advancement holds the potential to revolutionize the early detection and treatment of prostate cancer.

Key Findings

The study involved 61 readers, including 34 experts and 27 non-experts, from 17 countries. These readers assessed 360 MRI examinations of men with prostate cancer, with and without AI assistance. The AI system used was developed by the Prostate Imaging-Cancer AI (PI-CAI) Consortium, specifically for the detection and diagnosis of clinically significant prostate cancer (csPCa).

Methodology

The primary objective was to determine whether AI-assisted csPCa diagnosis was superior to unassisted diagnosis at the patient level. Researchers used the area under the receiver operating characteristic curve (AUROC), sensitivity, and specificity at a Prostate Imaging Reporting and Data System (PI-RADS) threshold of 3 or more as key metrics.

Among the 360 men examined, 122 had csPCa. The results were impressive: the AUROC was 0.916 with AI assistance compared to 0.882 without, showing a 3.3% improvement. Sensitivity improved by 2.5% (96.8% for AI-assisted vs 94.3% for unassisted), and specificity increased by 3.4% (50.1% for AI-assisted vs 46.7% for unassisted).

Non-Expert Readers Benefit Most

Notably, non-expert readers showed the greatest performance improvement with AI assistance. They achieved higher AUROC scores than those of unassisted experts, highlighting the potential of AI to level the playing field in medical diagnostics.

Practical Implications

The findings suggest that AI assistance can significantly improve csPCa diagnosis when compared with unassisted assessments of biparametric MRI. The improvements in AUROC, sensitivity, and specificity at a PI-RADS score of 3 or more are statistically significant. Non-expert readers, in particular, demonstrated higher benefits from AI assistance compared to expert readers.

Limitations and Future Directions

While the study's results are promising, it is important to note some limitations. The data were retrospectively curated within the PI-CAI scope, resulting in a mix of consecutive and sampled cohorts. The study's generalizability requires further validation across external cohorts with varying disease prevalence, image quality, and other clinical factors. Additionally, the controlled online reading workstation environment differed from readers' native settings, which may have affected diagnostic performance. The study did not assess workflow efficiency or the clinical applicability of performance improvements in real-world settings.

Funding and Disclosures

The study received funding support from Health-Holland and the European Union's Horizon 2020. Several authors reported receiving personal fees and research funding from various sources.

[Related: Future of AI in Healthcare]