Automating Colon Polyp Classification in Digital Pathology by Evaluation of a "Machine Learning as a Service" AI...

Area of Science:

• Digital Pathology
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• Artificial intelligence (AI) models are increasingly vital for enhancing pathological diagnoses.
• Cloud-based machine learning platforms offer readily available AI tools for domain experts like pathologists.
• This study explores the use of AutoML for developing AI in digital pathology.

Purpose of the Study:

• To evaluate AutoML algorithms for creating robust colon polyp classification models.
• To assess the applicability of these AI models in digital pathology workflows.

Main Methods:

• Utilized whole-slide images from public and institutional databases for training.
• Developed an AI model using Google's VertexAI AutoML platform.
• Classified colon polyps into hyperplastic polyps, tubular adenomas, and normal colon.

Main Results:

• The AI model achieved 100% accuracy for tubular adenoma and hyperplastic polyp classification.
• Normal colon tissue was classified with 97% accuracy.
• Demonstrated very low sensitivity and specificity error rates.

Conclusions:

• Accessible AutoML algorithms can effectively develop diagnostic AI models for digital pathology.
• AI models trained on whole-slide images can improve pathologists' diagnostic efficiency.
• This approach facilitates the integration of AI into routine pathology practice.