Exploring Variations in AI-Based Breast Cancer Risk Prediction: An Analysis of Pathology and Receptor Types

Purpose: Artificial intelligence (AI) algorithms have been developed to predict short-term breast cancer risk among patients. This study aimed to assess whether a commercially available image-based AI model predicts breast cancer risk differently across various breast cancer pathologies and receptor types.

Methods/Materials: A retrospective analysis was conducted on patients who underwent screening mammograms in a network of breast imaging clinics. Clinical features and pathology reports were collected from patients diagnosed with breast cancer in 2021. AI risk scores were analyzed based on screening mammograms from the year prior to diagnosis. Statistical analysis was performed using logistic regression in STATA.

Results: We identified 790 patients diagnosed with breast cancer in 2021. The average age was 65±12. Among them, 73% were white, 11% were black, 9% were Hispanic, 3% were Asian, and 5% were other or unknown race. In addition, 32% had ductal carcinoma in situ (DCIS), 58% had invasive ductal carcinoma, 7% had invasive lobular carcinoma, 2% had other invasive carcinoma, and 1% had lymph node metastasis only. Of the 506 patients with risk scores available from the prior year’s screening mammogram, the average score was 0.69±0.54. The average risk score was 0.66±0.56 for DCIS, 0.69±0.54 for invasive ductal carcinoma, 0.90±0.46 for invasive lobular carcinoma, 0.58±0.51 for other types of invasive carcinoma, and 0.3±0.17 for lymph node metastasis only. Regarding receptor types, the average risk score was 0.62±0.46 for ER and PR-positive cancer, 0.61±0.62 for triple-negative cancer, and 1.09±0.60 for HER-positive cancer.

Conclusions: Our data revealed varied risk scores the year before breast cancer diagnosis for different pathologies and receptor types. In our data, patients with invasive lobular carcinoma and HER-positive cancer had higher risk scores. By discerning variations in risk scores across different breast cancer pathologies and receptor types, physicians can potentially tailor screening and preventive strategies, fostering a more personalized approach to breast cancer care and improving patient outcomes.