Analysis of directional patterns of lung nodules in computerized tomography using Getis statistics and their accumulated forms as malignancy and benignity indicators

The large incidence of lung cancer in Brazil and around the world, in addition to its difficult diagnosis, especially in the initial stages, has been driving efforts to develop tools that support image-based diagnosis. The main objective is to avoid invasive procedures, which usually pose risks to patients. This work uses Getis spatial autocorrelation statistics, Getis∗, plus its accumulated forms to verify patterns occurring in geographic areas, aiming to indicate the nature of the lung nodule (benign or malignant). Nodule analysis is performed on its volume in a directional way, checking whether there are distances inside the nodule with large intensity variability of the voxels, for malignant and benign nodules. The classification is done by selecting the best four features from the 2400 generated features, for each of the Getis estimates. The Lung Image Database Consortium (LIDC) is used to verify the efficacy of the measures in the diagnosis. Results have shown that all of the Getis estimates succeeded in the discrimination of nodules in LIDC, with accuracy higher than 80% and confirmed by three different classifiers.

► This work presents a methodology for recognition of directional patterns of spatial distribution.
► We propose an innovative use of the Getis statistic and its variation to indicate the nature of the lung nodule (benign or malignant).
► The classifiers used in this work were support vector machine, nearest mean classifiers, and the linear classifier based on normal density.
► Results have shown that the accuracy obtained was higher than 80%.