Artificial intelligence in MRI for the detection and surveillance of prostate cancer
Artificial intelligence in MRI for the detection and surveillance of prostate cancer
Prostate cancer (PCa) remains a major global health burden, with 1.4 million new cases reported in 2020. While MRI plays a central role in detecting PCa, its diagnostic accuracy is highly dependent on radiologist expertise, leading to variability in outcomes.
This thesis of Christian Roest explores the potential of MRI-based artificial intelligence (AI) to enhance both the diagnosis and surveillance of PCa. Particular attention is given to active surveillance, where AI applications remain underexplored. We present novel approaches, including sequential AI models for tracking disease progression over time, and hybrid models that combine MRI data with clinical parameters to support noninvasive monitoring and personalized risk assessment. To further improve model accuracy, we investigate the integration of clinical data and techniques for standardizing multicenter MRI datasets. Additionally, we propose transfer learning strategies to overcome the limitations of small sequential imaging datasets, demonstrating improved performance in detecting clinically significant PCa. Overall, the findings highlight the promise of AI in delivering more accurate, consistent, and individualized prostate cancer care.