Congress:
ECR25
Poster Number:
C-24405
Type:
Poster: EPOS Radiologist (scientific)
Authorblock:
M. Wodrich, F. Sahlin, J. Karlsson, I. Arvidsson, K. Lang; Lund/SE
Disclosures:
Marisa Wodrich:
Nothing to disclose
Freja Sahlin:
Nothing to disclose
Jennie Karlsson:
Nothing to disclose
Ida Arvidsson:
Nothing to disclose
Kristina Lang:
Nothing to disclose
Keywords:
Artificial Intelligence, Breast, Oncology, Ultrasound, Comparative studies, Computer Applications-Detection, diagnosis, Screening, Cancer
- Radiologists' performance on POCUS and BUS was similar, with no significant difference.
- AI performed slightly better on BUS than on POCUS, with higher specificity than average breast radiologists on both data sets.
- AI outperformed the radiologists on BUS, with higher specificity for the same sensitivity.
- On POCUS, radiologists performed better than AI, which missed one malignant case.
Our results serve as a proof of concept and indicate the potential of using AI to automatically analyze POCUS breast images in limited-resource settings. Further studies with larger data sets and more readers are needed to better estimate the performance. Efforts are needed to improve the AI methods stability against noisy data and outliers.