Detection of Artifacts in Mammographic Images Using Deep Learning Techniques

Accurate detection and classification of artifacts in mammographic images is critical in diagnosing and following breast pathologies. Artifacts generated during mammogram capture and caused by software processing errors or detector problems can distort image interpretation and compromise diagnostic accuracy. This work implements artifact detection, specifically two artifacts related to the mammogram unit: Contrast Splatter (CS) and White Pixel (WP). A state-of-the-art convolutional neural network was used to detect and classify the data: YOLOV8. Tests were run in the modified database from CBIS-DDSM, where artifacts were synthetically added according to actual parameters reported in the bibliography. Results show promising results using YOLOV8, with a 0.95 and 0.97 recall index for CS y WP, respectively. However, other metrics show low accuracy and precision performance for WP detection. This work shows the usability of artificial intelligence methods in detecting problems with mammographic units that can lead to missing results on invalid studies in this important health topic.