Breast Cancer Diagnosis From Biopsy Images Using Generic Features and SVMs

AbstractA fully automatic method for breast cancer diagnosis based on microscopic biopsy images is presented.The method achieves high recognition rates by applying multi-class support vector machines on generic featurevectors that are based on level-set statistics of the images. We also consider the problem of classification withrejection and show preliminary results that point to the potential benefits.Index Termsbreast cancer, biopsy, automatic diagnosis, SVM, image processing, feature extraction I. I NTRODUCTION Recent years have witnessed a large increase of interest in automated and semi-automated breast cancerdiagnosis [1], [2], [3], [4], [5], [6], [7], [8], [9]; see [10] for a recent review. In this paper we present anautomatic classification method for breast cancer diagnosi s based on microscopic biopsy images. In particular,we consider the problem of classifying a tissue specimen as either healthy, tumor in situ 1 , or invasive carcinoma.We propose a fully automatic classification method, using ge neric features and state-of-the-art statisticallearning algorithms and methodologies. We experiment with a dataset that was previously used in [11], wherea highly specialized morphology-based feature generationprocess was used. We show here that simple genericfeatures lead to slightly superior (6.6% vs. 8.0%) error rate.A desirable functionality of automated or semi-automated medical diagnosis systems is the option of ‘decisionwith rejection’ whereby the system generates decisions with confidence larger than some prescribed thresholdand transfers the decision on cases with lower confidence to a human expert. In this work we also examine asimple method for decision with rejection and demonstrate its viability.Figure 1 presents three sample images of healthy tissue, tumor in situ and invasive carcinoma. Note that theimages are not clear-cut, and contain much information of nodiagnostic significance. These were the individualsamples in our work; they were not subdivided or cropped in any way. For more information on the dataset,see Section IV.