Clinical Aspects to Consider in Constructing a Standardized Test Set for Validation of Clinical Decision Support System for Mammography based on Artificial Intelligence

Abstract

In the era of artificial intelligence (AI), various computed aided detection (CAD) algorithms are being developed for mammography, and the need for a standardized test set that can be generally applied for clinical validation has been voiced. During the process of constructing a standardized test set, an expert meeting was held in Seoul, Korea at June 2018, in order to reach consensus on factors to consider in building this test set for validating AI-based CAD tools developed for mammography. Faculty invited for the consensus meeting were dedicated breast radiologists, who were considered experts in this field. Topics discussed during the meeting were population-related factors, consensus on final diagnosis and interpreting lexicons according to the ACR BI-RADS, and the level of difficulty of the mammography for interpretation. Results of the discussion are summarized in this report.In the era of artificial intelligence (AI), various computed aided detection (CAD) algorithms are being developed for mammography, and the need for a standardized test set that can be generally applied for clinical validation has been voiced. During the process of constructing a standardized test set, an expert meeting was held in Seoul, Korea at June 2018, in order to reach consensus on factors to consider in building this test set for validating AI-based CAD tools developed for mammography. Faculty invited for the consensus meeting were dedicated breast radiologists, who were considered experts in this field. Topics discussed during the meeting were population-related factors, consensus on final diagnosis and interpreting lexicons according to the ACR BI-RADS, and the level of difficulty of the mammography for interpretation. Results of the discussion are summarized in this report.