BONEXAI-NET: AN EXPLAINABLE ARTIFICIAL INTELLIGENCE FRAMEWORK FOR BONE TUMOR CLASSIFICATION AND SURVIVAL PREDICTION

Abstract

Bone tumors are one such varied class of benign and malignant tumors, which bring about numerous challenges in terms of their clinical diagnosis, prognosis, and therapy. Classification and survival prediction are critical aspects to consider when improving patients' conditions and providing individualized treatment strategies. Conventional classification techniques are heavily based on expert opinions derived from the clinical and pathological presentation of the disease, which can be time-consuming and highly variable. The current study is aimed at developing a new solution in the form of BoneXAI-Net for bone tumor classification and survival prediction. In the proposed methodology, the use of comprehensive data pre-processing, feature engineering, and machine learning based predictive models will help in determining the key factors that are related to the tumor and survival of the patients. A number of different classifiers such as Random Forest, XGBoost, LightGBM, and CatBoost have been used and their outputs have been fused together using the optimization based ensemble learning method. Explainable AI approaches like SHAP will be used for increasing interpretability. Experiment evaluation is done on a publicly available bone tumor dataset that contains demographic, clinical, pathological, and treatment features. Performance is evaluated in terms of accuracy, precision, recall, F1 score, and area under receiver operating characteristic curve (AUC-ROC). The results show that BoneXAI-Net outperforms other baseline models on classification accuracy and survival prediction. Moreover, the explainable machine learning aspect helps doctors understand what affects the prediction so that they can make decisions based on that knowledge in orthopedic oncology. The proposed framework will play an instrumental role in the development of precision medicine in that it combines high prediction power with transparency and therefore offers a credible and clear clinical support system for diagnosing bone tumors and managing patients.