Breast Density Classified Using AI-aided Mammographic Breast Density Classification Tool and Validated among Hilla Women's Sample

Background: Breast density evaluation is very important in breast cancer screening, since it affects both cancer risk and the sensitivity of mammography detection. Radiologists' conventional visual judgements could differ greatly among readers. Automated software driven by machine learning has been created to examine mammographic pictures and generate consistent breast density classifications in order to solve this. This work intends to test the degree of agreement between conventional visual interpretation by radiologists and artificial intelligence (AI)-based computerised breast density evaluation. Methods: A total of 300 digital mammography images from women aged 40 to 72 years were analyzed. Two experienced radiologists assessed breast density independently, according to the BI-RADS 5th Edition classification. Additionally, the images were processed using an AI-based software (Lunit INSIGHT MMG version 1.1.8.0) for automatic classification. Inter-observer agreement was evaluated using kappa (κ) statistics, while reliability among readings was assessed using Cronbach’s alpha. Results: Substantial agreement was found between the two radiologists (κ = 0.69). Similarly, substantial agreement was observed between the first radiologist and the AI model (κ = 0.61), and the second radiologist and the AI model (κ = 0.60). Reliability testing showed good internal consistency across all three readings and between each radiologist and the AI. Overall, 78% of cases had concordant density classifications among the three readings. There was no significant difference in age between concordant and discordant cases. Conclusions: The AI model demonstrated a substantial level of agreement with human radiologists, supporting its potential as a reliable tool in routine breast density assessment.