Análise temporal de lesões pulmonares através de textura e forma dinâmicas

Abstract

Lung cancer is still one of the most popular incidents worldwide. Late detection of neoplasia and the proliferation capacity of malignant cells hinder targeted therapies, which often fail. The temporal evaluation is a useful tool to analyze the biological behavior of a lung lesion before, during and after treatment, or lesions of undetermined diagnosis. This paper aims to provide more detailed information about the changes of lesions complementing studies on the biological activity of lung masses and nodules. The methodology presented describes changes in texture and shape of a lung lesion over time. The lesion tissues are segmented using a mixture of dynamic texture model, according to the texture change over time. Texture change descriptors are extracted from each segmented region, while shape change descriptors are extracted from the entire lesion. The study was conducted on two bases of chest-computed tomography: a Public Lung Database, which has lesions that undergo evaluation for drug therapy; and a private base of initially indeterminate diagnosis of lung lesions, but later classified as benign after lesion follow-up. The results of the proposed techniques showed that the variance in Z-score of measures extracted from shape change in the lesions was 0.02 in the private database and 1.47 in the public database, while the variance of measures extracted from texture change was 0.49 in the private database and 1.15 in the public database. A classification model was proposed with the descriptors extracted to predict the diagnosis of the lesion, resulting in an accuracy of 97.2%.