Computer-aided Diagnosis (CADx) of breast imaging deals with the application of computer tools to provide recommendations regarding the malignancy status and clinical management of breast lesions detected during mammographic screening.  Typically a CADx system is expected to help radiologists determine if an already detected lesion requires biopsy to determine its malignancy status.  Reducing the number of unnecessary biopsies is an important clinical task as the majority (65%-85%) of biopsies performed due to suspicious mammograms are found to be benign.  The economic cost, physical burden, and emotional stress associated with excessive biopsy of benign lesions have been reported before.  Furthermore, another well-documented problem is the variability among radiologists regarding the recommended clinical management (biopsy vs. short-term follow-up) of suspicious breast lesions.  CADx systems aim to improve the sensitivity, specificity, efficiency, and cost-effectiveness of breast cancer screening programs.

Fig 1. Flow chart of typical CADx system

Several RAILab members (Drs. Baker, Bilska-Wolak, Floyd, Lo, Tourassi) have been actively pursuing the development and clinical evaluation of CADx algorithms in both mammographic and sonographic breast imaging.  The algorithms are designed to capitalize on various sources of clinical information: computer-extracted image features, radiologist-interpreted findings, and patient history findings. The features are carefully selected and merged together using a classifier (Fig. 1).  Our CADx classifiers span a wide range of decision algorithms based on machine learning (e.g., artificial neural networks, case-based reasoning, knowledge-based analysis) and statistical modeling (e.g., linear, likelihood ratio-based, and information-theoretic classifiers). The algorithms are tailored to address issues of accuracy, interpretability, and robustness.

In a recent study on the independent clinical evaluation of our case-based reasoning and likelihood-ratio based CADx systems, both systems were able to correctly identify 98% of malignant masses while sparing almost half of the benign masses from unnecessary biopsy.  To enhance these predictive models with interpretability, we are also pursuing an information-theoretic content-based image retrieval (CBIR) approach as the basis of a CADx system.  Given a query lesion, the CBIR algorithm retrieves similar cases from a knowledge databank.  The retrieved images are presented to the radiologist in the rank order according to similarity with their corresponding malignancy status (Fig. 2).  In this manner, radiologists are given visual justification of the computer’s recommendation.