Comprehensive Assessment of MRI-based Artificial Intelligence Frameworks Performance in the Detection, Segmentation,...

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Traditional interpretation of prostate MRI for cancer detection faces challenges due to reader variability.
• Artificial intelligence (AI) offers potential solutions for automated and semi-automated analysis of prostate MRI.
• Open-source datasets are crucial for developing and validating diverse AI frameworks.

Purpose of the Study:

• To conduct an in-depth performance assessment of MRI-based AI frameworks for prostate cancer.
• To evaluate AI capabilities in detecting, segmenting, and classifying prostate lesions.
• To analyze AI performance using publicly available, open-source MRI databases.

Main Methods:

• Systematic review and in-depth assessment of MRI-based AI frameworks.
• Utilized open-source MRI datasets specifically curated for prostate cancer research.
• Included 17 datasets, with 12 focused on prostate cancer detection/classification.
• Analyzed 52 studies that met the inclusion criteria for the assessment.

Main Results:

• AI frameworks demonstrate varying performance in detecting, segmenting, and classifying prostate cancer lesions on MRI.
• Open-source datasets provide a valuable resource for evaluating the generalizability of AI tools.
• The study identified key areas where AI shows significant potential and areas needing further development.

Conclusions:

• MRI-based AI frameworks hold significant potential to enhance the accuracy and consistency of prostate cancer diagnosis.
• Further research and validation using diverse open-source datasets are essential for clinical translation.
• AI can help mitigate interreader variability in prostate MRI interpretation, leading to improved patient outcomes.