iCAP: An Individualized Model Combining Gaze Parameters and Image-Based Features to Predict Radiologists' Decisions...

Area of Science:

• Radiology
• Medical Imaging
• Human-Computer Interaction

Background:

• Mammography interpretation is complex and prone to errors, impacting patient outcomes.
• Identifying subtle signs of breast cancer requires meticulous analysis of mammograms.
• Current interpretation methods can lead to both missed cancers (false negatives) and unnecessary biopsies (false positives).

Purpose of the Study:

• To introduce and evaluate an individualized tool, eye-Computer Assisted Perception (iCAP), for identifying mammogram interpretation errors.
• To assess iCAP's effectiveness in improving the accuracy of cancer detection and reducing false positives.
• To leverage eye-tracking data and image features for enhanced mammogram analysis.

Main Methods:

• Developed iCAP with two modules: one for marked suspicious areas (TP/FP) and another for fixated unmarked areas (TN/FN).
• Collected eye-tracking data from eight radiologists reading 120 digital mammograms (59 with biopsy-proven cancer).
• Built user-specific support vector machine models using gaze parameters and image features, validated with leave-one-out cross-validation.

Main Results:

• Retrospective testing in a simulated scenario showed iCAP's potential.
• Average increase of 12%±6% in correctly localized cancer.
• Average decrease of 44.5%±22.7% in false positives per image.

Conclusions:

• iCAP demonstrates significant potential to enhance mammogram interpretation accuracy.
• The tool effectively reduces both false positives and improves cancer localization.
• Individualized analysis of radiologist eye-tracking and image data offers a promising approach for computer-assisted perception in mammography.