Medical Image Classification via SVM using LBP Features from Saliency-Based Folded Data

摘要

Good results on image classification and retrieval using support vectormachines (SVM) with local binary patterns (LBPs) as features have beenextensively reported in the literature where an entire image is retrieved orclassified. In contrast, in medical imaging, not all parts of the image may beequally significant or relevant to the image retrieval application at hand. Forinstance, in lung x-ray image, the lung region may contain a tumour, hencebeing highly significant whereas the surrounding area does not containsignificant information from medical diagnosis perspective. In this paper, wepropose to detect salient regions of images during training and fold the datato reduce the effect of irrelevant regions. As a result, smaller image areaswill be used for LBP features calculation and consequently classification bySVM. We use IRMA 2009 dataset with 14,410 x-ray images to verify theperformance of the proposed approach. The results demonstrate the benefits ofsaliency-based folding approach that delivers comparable classificationaccuracies with state-of-the-art but exhibits lower computational cost andstorage requirements, factors highly important for big data analytics.