CT Texture Analysis and Machine Learning Improve Post-ablation Prognostication in Patients with Adrenal Metastases: A Proof of Concept.

Abstract

To assess the performance of pre-ablation computed tomography texture features of adrenal metastases to predict post-treatment local progression and survival in patients who underwent ablation using machine learning as a prediction tool. This is a pilot retrospective study of patients with adrenal metastases undergoing ablation. Clinical variables were collected. Thirty-two texture features were extracted from manually segmented adrenal tumors. A univariate cox proportional hazard model was used for prediction of local progression and survival. A linear support vector machine (SVM) learning technique was applied to the texture features and clinical variables, with leave-one-out cross-validation. Receiver operating characteristic analysis and the area under the curve (AUC) were used to assess performance between using clinical variables only versus clinical variables and texture features. Twenty-one patients (61% male, age 64.1 ± 10.3years) were included. Mean time to local progression was 29.8months. Five texture features exhibited association with progression (p < 0.05). The SVM model based on clinical variables alone resulted in an AUC of 0.52, whereas the SVM model that included texture features resulted in an AUC 0.93 (p = 0.01). Mean overall survival was 35months. Fourteen texture features were associated with survival in the univariate model (p < 0.05). While the trained SVM model based on clinical variables resulted in an AUC of 0.68, the SVM model that included texture features resulted in an AUC of 0.93 (p = 0.024). Pre-ablation texture analysis and machine learning improve local tumor progression and survival prediction in patients with adrenal metastases who undergo ablation.