Predicting pathological staging of non-small cell lung cancer using a multi-task radiomics model integrating intratumoral and peritumoral features.

Pathological staging is essential for guiding treatment decisions and determining prognosis in patients with non-small cell lung cancer (NSCLC). The present study aimed to establish a model using intratumoral and peritumoral features from computed tomography radiomics data and a multi-task learning algorithm, and evaluate its predictive performance for the pathological stage of NSCLC. Data from 198 eligible patients with NSCLC from The Cancer Imaging Archive database were retrospectively analyzed, which was used to develop four radiomics models to classify the pathological stages of NSCLC. These models combined the traditional random forest and multi-task random forest algorithms with the volumes of interest for the intratumoral region alone and with the intratumoral and peritumoral regions combined. Subsequently, the data from 90 patients from a real-world dataset were collected for use as an external test set. Diagnostic performance was evaluated using accuracy, precision, sensitivity, specificity, F1 scores and receiver operating characteristic curves. The results revealed that, in the internal test set, the area under the curve (AUC) values for model 1 (single-task model based on the intratumoral region), model 2 (single-task model combining intra- and peritumoral regions), model 3 (multi-task model based on the intratumoral region) and model 4 (multi-task model combining intra- and peritumoral regions) were 0.814, 0.900, 0.896 and 0.938, respectively. In the external test set, the AUC values were 0.821, 0.921, 0.858 and 0.939, separately. Moreover, the results of the DeLong test indicated that the AUC difference between model 1 and 4 was statistically significant (P<0.05). In conclusion, the multi-task radiomic model incorporating both intratumoral and peritumoral regions demonstrated favorable diagnostic efficacy in the predictive pathological staging of NSCLC.