Development of a tool for automatic classification of intratumoral heterogeneity of lung cancers based on PET/CT intensity values

Intratumoral heterogeneity, typical of solid tumors, results from the coexistence of various subpopulations of tumor cells with different biological characteristics. The objective of this work was to develop an algorithm for the analysis and classification of intratumoral heterogeneity in lung cancers. The classification algorithm was based on features obtained from PET/CT images of the tumors. Delimited regions containing the tumors were defined by experience observers, who also classified each tumor regarding its degree of heterogeneity using a scale of four levels for heterogeneity. Features were created based on the joint histogram built from CT and PET data, which were then used to implement two classifier algorithms: one ordinal regression exploiting the four levels of heterogeneity and a logistic regression checking only two. Statistical analysis was carried out using IBM-SPSS Statistics 20 with a level of significance of 0.05. The models developed show good accuracy, mainly the binary classification algorithm (90%).