Impact of image quality and clinical radiographic features on AI model for detecting clinically significant prostate cancer: a multicenter study

Motivation: The widespread clinical adoption of AI in prostate cancer diagnosis is hindered by several factors, limiting its generalization and broader use. Goal(s): To investigate the impact of image quality and clinical radiographic features on the performance of AI model in detecting clinically significant prostate cancer (csPCa). Approach: The researchers proposed several parameters for image quality and radiographic features, comparing these parameters between patients correctly diagnosed by the AI model and those with incorrect diagnoses. Results: The diagnostic accuracy of the AI model is notably greater for patients with more cancerous lesions, larger total lesion volumes, and lower average ADC values. Impact: It identifies the factors influencing the generalization performance of prostate AI, thereby enhancing clinicians' understanding and confidence in its application, ultimately enabling more effective use in practice.