Initial interactions with the FDA on developing a validation dataset as a medical device development tool.

Area of Science:

• Computational pathology
• Digital pathology
• Biomedical imaging analysis

Background:

• Quantifying tumor-infiltrating lymphocytes (TILs) in breast cancer is crucial but challenging for pathologists.
• Whole slide imaging enables computational models for TIL quantification, but their development is resource-intensive.
• Standardized validation is needed for computational models used in breast cancer diagnostics.

Purpose of the Study:

• To propose a dataset for validating computational models that quantify TILs in breast cancer.
• To reduce the regulatory burden for developers of TIL quantification models through FDA's Medical Device Development Tool (MDDT) program.
• To facilitate open, fair, and consistent performance evaluation of computational models.

Main Methods:

• Preparation and submission of a dataset to the FDA's MDDT program.
• Engaging with the FDA to understand requirements for qualifying validation datasets.
• Discussing the MDDT process and its implications for computational model validation.

Main Results:

• The study outlines the process of preparing and submitting a dataset for MDDT qualification.
• It details initial feedback received from the FDA regarding the submission.
• The proposed qualified dataset aims to enable head-to-head comparisons of multiple computational models.

Conclusions:

• A qualified MDDT validation dataset can standardize and simplify the regulatory process for computational pathology tools.
• This approach can foster trust and consistency in the performance assessment of TIL density estimation models.
• Sharing experiences with the FDA MDDT process aids the broader community in developing and validating AI-driven pathology tools.