Identification of error making patterns in lesion detection on digital breast tomosynthesis using computer-extracted image features
Digital breast tomosynthesis (DBT) can improve lesion visibility by eliminating the issue of overlapping breast tissue present in mammography. However, this new modality likely requires new approaches to training. The issue of training in DBT is not well explored. We propose a computer-aided educational approach for DBT training. Our hypothesis is that the trainees' educational outcomes will improve if they are presented with cases individually selected to address their weaknesses. In this study, we focus on the question of how to select such cases. Specifically, we propose an algorithm that based on previously acquired reading data predicts which lesions will be missed by the trainee for future cases (i.e., we focus on false negative error). A logistic regression classifier was used to predict the likelihood of trainee error and computer-extracted features were used as the predictors. Reader data from 3 expert breast imagers was used to establish the ground truth and reader data from 5 radiology trainees was used to evaluate the algorithm performance with repeated holdout cross validation. Receiver operating characteristic (ROC) analysis was applied to measure the performance of the proposed individual trainee models. The preliminary experimental results for 5 trainees showed the individual trainee models were able to distinguish the lesions that would be detected from those that would be missed with the average area under the ROC curve of 0.639 (95% CI, 0.580-0.698). The proposed algorithm can be used to identify difficult cases for individual trainees.
