Online Poster Portal

  • Author
    Kathleen Kilroe
  • Discovery PI

    Dr. Hannah Milch, MD

  • Project Co-Author

  • Abstract Title

    Leveraging AI to Reduce the Need for Prior Mammograms in Low-Risk Patients: A Step Toward Efficient and Cost-Effective Breast Cancer Screening

  • Discovery AOC Petal or Dual Degree Program

    Health Justice & Advocacy

  • Abstract

    Title: Leveraging AI to Reduce the Need for Prior Mammograms in Low-Risk Patients: A Step Toward Efficient and Cost-Effective Breast Cancer Screening

    Author: Kathleen Kilroe

    Area of Concentration (Petal): Research, Health Equity & Advocacy Petal

    Specialty (if any): Diagnostic Radiology

    Keywords: Artificial Intelligence, Breast Mammography, Clinical Efficacy

    Background: It is standard practice to use prior mammograms, when available, to aid in current BI-RADS classification. While beneficial, this adds cost, delays, and radiologist workload. With AI—particularly deep learning—proving effective in lesion detection and BI-RADS accuracy, there’s potential to reduce reliance on priors. We propose that for patients without readily available prior imaging, a ‘low’ risk designation by AI could eliminate the need to retrieve earlier mammograms.

    Objective: Evaluate the efficacy of AI interpretation of 3D mammography in accurately designating patients without prior mammograms available (BI-RADS 0) as ‘low’ risk.

    Methods: Analysis uses data from an IRB-approved pragmatic RCT at UCLA that compares radiologist performance and patient outcomes when 3D mammograms are evaluated with versus without the FDA-approved AI tool, Transpara. For patients graded ‘BI-RADS 0: prior images needed’ by the radiologist and graded “low” risk by Transpara, we will explore imaging outcomes to identify if AI screening may replace the need to evaluate prior mammograms in low-risk patients. Data analysis includes descriptive analysis, Chi square, and ANOVA.

    Results: Among BI-RADS 0 patients needing priors and labeled “low” risk by Transpara (N=468), 69% were downgraded to a lower-risk BIRADS after receiving priors, while 31% were re-assigned to ‘BI-RADS 0: needs more imaging’. After more imaging, 77% were downgraded to lower-risk BIRADS, and 17% were upgraded to BIRADS 4 (biopsy), of which 85% were benign, 10% high risk (Atypical Lobular Hyperplasia), and 5% malignant (Ductal carcinoma in-situ).

    Conclusions: AI accurately identified true low-risk patients without prior mammograms. Most were downgraded after receiving priors or additional imaging, and among those biopsied, the majority were benign or high-risk, with only one malignant case. AI may reduce the need for priors in low-risk patients, improving screening efficiency.