Deep Learning Methodologies Applied to Digital Pathology in Prostate Cancer: A Systematic Review

Area of Science:

• Digital Pathology
• Artificial Intelligence
• Oncology

Background:

• Digital pathology enables the use of artificial intelligence (AI), specifically deep learning (DL), for analyzing digitized slides.
• Prostate cancer (PCa) is a key area where DL applications are being explored in pathology.

Purpose of the Study:

• To systematically review DL applications and their performance in PCa within digital pathology.
• To assess the quality and identify biases in existing DL studies for PCa.

Main Methods:

• Systematic literature search of PubMed and Embase databases.
• Risk of Bias (RoB) assessment using an adapted QUADAS-2 tool.
• Categorization of studies into pre-processing, diagnosis, and prediction tasks.

Main Results:

• Out of 77 studies, 8 focused on pre-processing, 53 on diagnosis (cancer detection, Gleason grading), and 15 on prediction (recurrence, genomics).
• DL achieved high performance in cancer detection, with Area Under the Curve (AUC) up to 0.99, with some algorithms nearing routine use.
• Common biases identified include a lack of external validation in many studies.

Conclusions:

• DL is a powerful tool for PCa analysis in digital pathology, particularly for cancer detection.
• While DL shows promising results, addressing biases like the need for external validation is crucial for clinical translation.
• Further research should focus on robust validation to ensure the reliability of DL algorithms in routine PCa diagnosis and prediction.