An Argument for MLOps Pipelines for AI-Based Gleason Scoring Systems

Gleason scores are the gold standard for grading prostate cancer. However, human variability can affect the score; that is, manual scoring can produce a grade that is not reflective of the cancer. This variability can result in over- or undertreatment, which can lead to metastasis, recurrence, or harmful overtreatment. AI (Artificial Intelligence) in the form of machine vision is currently being researched to create systems that can limit variability; however, AI models are imperfect and can lose accuracy for multiple reasons. Compounding the issue, retraining can be an in-depth and labor-intensive endeavor. To streamline retraining, MLOps pipelines can be used to recalibrate and reverify AI-Based Gleason grading systems. This article argues that MLOps pipelines are essential to ensure long-term clinical reliability of AI-based Gleason grading systems and proposes a practical, adaptable architecture to combat model drift amid evolving oncology practices.