MUFASA

Information utility-aware preprocessing for reliable computational pathology model reasoning

MUFASA is an information utility-aware preprocessing framework designed to improve the reliability of model reasoning in computational pathology. The project focuses on improving whole-slide image data quality before downstream machine learning analysis.

The central goal is to remove or reduce the influence of non-diagnostic and low-utility image regions while preserving diagnostically meaningful tissue patterns for reliable downstream prediction.

Role: Collaborating researcher

Key contributions:

  • Supported development of an AI-powered preprocessing framework for whole-slide image dataset quality improvement.
  • Contributed to pathology image preprocessing workflows involving artifact filtering, tile extraction, and quality control.
  • Helped evaluate preprocessing strategies for improving model reasoning in computational pathology.
  • Contributed to research dissemination through Stanford Pathology Research Retreat and manuscript preparation.

Research relevance:

  • Artifact-aware WSI preprocessing
  • Dataset quality control
  • Reliable model reasoning
  • Computational pathology
  • Whole-slide image machine learning

Related work:

Rathinaraja Jeyaraj, Barathi Subramanian, Jeanne Shen, et al., “MUFASA: An Information Utility-Aware Preprocessing Framework for Reliable Model Reasoning in Computational Pathology,” in preparation.