Machine Learning-Based Computed Tomography Texture Analysis of Lytic Bone Lesions Needing Biopsy: A Preliminary Study

Abstract

Introduction:Currently, medical imaging has a limited capacity to achieve a final histopathological diagnosis of bone lesions. This study aimed to evaluate the use of machine learning (ML)-based computed tomography (CT) texture analysis to determine benign and malignant behaviors of lytic bone lesions needing a biopsy.Methods:This retrospective study included 58 patients with lytic bone lesions. Lesion segmentation was independently performed by two observers. After evaluating unenhanced CT images, a total of 744 texture features were obtained. Reproducibility analysis and feature selection were used for dimension reduction. A training data set with a nested cross-validation approach was used for feature selection, optimization, and validation. Testing was executed on the remaining unseen data set. Classifications were done using five base ML classifiers and three voting strategies.Results:The best predictive performance was achieved using the k-nearest neighbors algorithm with six features. The area under the curve, accuracy, sensitivity, and specificity of the best algorithm were, respectively, 0.774%, 78.1%, 78%, and 78.1% for the validation data set; and 0.861, 82.4%, 82.4%, and 81.5% for the unseen test data set.Conclusion:The ML-based CT texture analysis may be a promising non-invasive technique for determining benign and malignant behaviors of lytic bone lesions that need a biopsy.

Keywords

• bone
• texture analysis
• radiomics
• machine learning
• artificial intelligence