Machine learning-based CT radiomics features for the prediction of pulmonary metastasis in osteosarcoma

• DOI: 10.1259/bjr.20201391
• Publication Date: 2021-06-11T07:13:10Z
• Authors: Helcio Mendonça P...; Maria Eugenia Lei...; Igor Ribeiro Dama...; Luiz Afonso de Ol...; Marcello Henrique...
• Keywords: Adult; Male; Osteosarcoma; Lung Neoplasms; Adolescent; Bone Neoplasms; Middle Aged; 3. Good health; Machine Learning; Young Adult; 03 medical and health sciences; 0302 clinical medicine; Predictive Value of Tests; Child, Preschool; Humans; Female; Child; Tomography, X-Ray Computed; Retrospective Studies

This study aims to build machine learning-based CT radiomic features predict patients developing metastasis after osteosarcoma diagnosis.This retrospective has included 81 with a histopathological diagnosis of osteosarcoma. The entire dataset was divided randomly into training (60%) and test sets (40%). A data augmentation technique for the minority class performed in set, along feature's selection model's training. were extracted from CT's image local Three frequently used learning models tried lung metastases (MT) those without (non-MT). According higher area under curve (AUC), best classifier chosen applied testing set unseen provide an unbiased evaluation final model.The predicting MT non-MT groups Random Forest algorithm. AUC accuracy results bulky (accuracy 73% [ 95% coefficient interval (CI): 54%; 87%] 0.79 [95% CI: 0.62; 0.96]). Features that fitted model (radiomics signature) derived Laplacian Gaussian wavelet filters.Machine radiomics approach can non-invasive method fair predictive risk pulmonary patients.Models based on analysis help assess osteosarcoma, allowing further studies worse prognosis.