Harmonization of features extraction process using the ComBat tool in a multicenter study on stage III unresectable Non-Small-Cell Lung Cancer.

In the BlueSky study we aim to explore the influence of different image acquisition parameters on radiomic features reproducibility in a series of stage III unresectable NSCLC patients undergoing chemo-radiotherapy and immunotherapy and to elaborate methods for correction and harmonization. The primary lung tumor was identified by oncological imaging specialists on two computed tomography scans acquired before and after chemoradiation. We extracted 42 radiomic features through the LIFEx software by applying a preprocessing image standardization phase. To assess the impact of different acquisition parameters on features extraction, we used the Combat tool with nonparametric adjustment and its longitudinal version LongComBat. Using the Kruskal-Wallis test on different acquisition parameters (scanner, KVp, convolution kernel, contrast agent, exposure time), we defined 14 acquisition protocols for the harmonization process. Before harmonization, $76%$ of the features were significantly influenced by these protocols. After, all the ComBat harmonized features resulted in being independent of the technical parameters, while $5%$ of the LongComBat ones still depended on acquisition protocols.