ROLE OF ARTIFICIAL INTELLIGENCE IN EVALUATION OF IMAGE READING OF MAMMOGRAM FOR CHARACTERIZATION OF BENIGN AND MALIGNANT LESIONS TAKING HISTOPATHOLOGY AS GOLD STANDARD

Abstract

Objectives: To assess the accuracy of AI software in the evaluation of mammograms for benign and malignant breast lesions as compared to Histopathology as the gold standard diagnosis.

Study design: Cross-sectional validation study

Study Setting and Duration: Azra Naheed Medical College Lahore, Pakistan, from April 2025 till June 2025.

Methodology: This study enrolled 210 women aged 25 to 65 years who presented to the hospital with breast-related symptoms using nonprobability consecutive sampling. The full-field digital mammograms these participants underwent were analyzed by AI software (Lunit INSIGHT MMG), which assigned a probability score for malignancy based on a cut-off of 59%. All participants underwent biopsy (core needle) to determine the histopathology results. Data analysis was completed using SPSS software, version 25, and included the calculation of sensitivity, specificity, and predictive values.

Results: 210 participated in this study with a mean age of 43.36 years (±9.19) and a mean BMI of 25.30 kg/m² (±3.01). The AI mammogram detected malignancies in 58.1% of the participants, which is comparable to the histopathology findings of 59.5%. The presence of malignant lesions was found to be significantly associated with a positive family history, smoking status, and marital status (all with p < 0.05).

Conclusions: The AI mammogram demonstrated a nearly accurate diagnosis of tumors that are in strong agreement with histopathology.