Histopathologic Basis for a Chest CT Deep Learning Survival Prediction Model in Patients with Lung Adenocarcinoma

This study investigates a preoperative CT-based deep learning (DL) model to predict disease-free survival in lung adenocarcinoma patients, providing histopathologic evidence. Data from 1667 patients who underwent curative resection without neoadjuvant therapy were collected. Seven histopathologic risk factors were recorded: aggressive adenocarcinoma subtype, mediastinal nodal metastasis, lymphatic, venous, and perineural invasion, visceral pleural invasion (VPI), and EGFR mutation status. Using 80 DL model-driven CT features, unsupervised clustering, and regression analyses, associations between patient clusters and histopathologic features were examined. Tumor annotations were created with AVIEW software (Coreline Soft) for generating ground truth data. The results showed that clusters 3 and 4 were associated with most histopathologic risk factors. The DL model output was significantly associated with aggressive adenocarcinoma subtype, venous invasion, and VPI, independently of semantic CT features. This study demonstrates the DL model's feasibility in identifying patients with histopathologic risk factors.