1814. Short-Time Improvement in Breast Positioning Performances Using a Dedicated AI-Platform for Real-Time Feedback on Mammography
Authors* Denotes Presenting Author
Friederike Riehle;
University Hospital Basel
Patryk Hejduk;
University Hospital Zurich
Karol Borkowski;
University Hospital Zurich
Silke Potthast;
Spital Limmattal
Noemi Schmidt *;
University Hospital Basel
Objective:
Diagnostic quality is well known to affect cancer detectability in mammography. However, regular quality
verification of the breast positioning in mammographic imaging is a challenge in the daily routine.
The aim of this study is to evaluate the added value of providing real-time feedback to the radiographers via
a commercial software platform based on artificial intelligence (AI) for the automatic determination of the
image quality following the perfect, good, moderate, and inadequate (PGMI) criteria.
Materials and Methods:
In this study, 3600 mammograms from our institution acquired in 2021 were analyzed. The quality of the breast positioning was assessed using a commercial AI software platform (b-box version 1.1, b-rayZ AG, Schlieren, Switzerland) for each breast. Craniocaudal (CC) and mediolateral oblique (MLO) projections were examined, according to the PGMI criteria. To access the impact of the software on the acquired image quality, 1205 images were evaluated before the integration and 2395 images were evaluated 3 months after the integration of the software into the daily routine.
Results:
After 3 months of software usage, the percentage of examinations considered perfect increased from 40% to 60%. The percentage of images considered inadequate decreased from 3% to 0.6%. The degree of improvement depended on the professional experience of the radiographer. With the help of the real-time visual feedback, especially less experienced radiographers, with fewer than 50 mammograms per year were able to improve the image quality. Moderate/inadequate images dropped from 9% to 3%. Additionally, the acquired image quality of more experienced radiographers improved by reducing moderate/inadequate images from 18% to 8%. Results are proven to be statistically significant with a p-value of 0.002.
Conclusion:
Automated and real-time quality assurance of mammograms in the daily routine significantly improved the image quality at the institution. The radiographer’s professional development, the institution’s quality standards, and documentation can be easily assessed and tracked with a good software platform. AI-driven software platforms can assist radiographers working in breast imaging by improving the positioning of the breast, resulting in significantly higher-quality images within short-term usage.
