Purpose: Osteosarcoma is a highly malignant bone tumor, and neoadjuvant chemotherapy (NACT) is commonly used to reduce tumor size before surgical resection. However, predicting the efficacy of NACT remains a challenge, as traditional clinical and imaging markers often lack sufficient sensitivity and specificity. Advanced imaging techniques like intravoxel incoherent motion (IVIM) and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) can evaluate tumor microenvironment characteristics, such as perfusion and diffusion properties. Recent studies suggest that habitat analysis, which segments the tumor into distinct regions with similar characteristics, may enhance predictive accuracy for treatment response. This study aims to evaluate the predictive value of habitat-based radiomic features extracted from IVIM and DCE-MRI in assessing NACT response in osteosarcoma patients, with the goal of identifying functional imaging biomarkers to enable more personalized treatment strategies.

Patients and Methods: Seventy-one patients with histologically confirmed osteosarcoma who underwent pre-treatment IVIM and DCE-MRI were retrospectively enrolled. Patients were classified into good and poor response groups based on histopathologic tumor necrosis rate following NACT. Whole-lesion histogram features and intratumoral habitat features were extracted from IVIM-Bi, IVIM-Mono, and DCE-MRI parametric maps. Feature selection included univariate analysis, correlation analysis, and stepwise logistic regression. Model performance was evaluated using ROC analysis, calibration curves, and decision curve analysis (DCA).

Results: No statistically significant whole-lesion histogram features were identified from the IVIM-Bi model. However, one feature from the IVIM-Mono model and two from the DCE-MRI model were retained. Habitat analysis identified three consistently present subregions (habitats 1, 2, and 5), from which 168 features were initially extracted. Following feature selection, four features from the IVIM-Bi model (HAB-IVIM-Bi) were retained and used to construct the final predictive model. The HAB-IVIM-Bi model achieved the highest performance among all models, with an AUC of 0.797, accuracy of 0.761, sensitivity of 0.719, and specificity of 0.795. DCA and calibration analysis confirmed its superior clinical utility and predictive reliability. 

Conclusions: Habitat analysis based on the IVIM-Bi model significantly outperformed histogram-based models and demonstrated strong predictive value for assessing NACT response in osteosarcoma. This approach captures intratumoral heterogeneity more effectively and holds promise for facilitating individualized treatment planning.