Characterization of Intratumoral Heterogeneity via MRI-Based Radiomic Habitats in Osteosarcoma

Accurate risk stratification for osteosarcoma is hindered by intratumoral heterogeneity. Conventional radiomics often treats tumors as homogeneous entities, overlooking spatial subregions and limiting prognostic accuracy.

To evaluate the prognostic value of an MRI-based radiomic habitat approach-partitioning the tumor into biologically distinct subregions-for post-treatment recurrence in osteosarcoma, comparing its performance with conventional whole-tumor analysis.

Retrospective cohort study.

Eighty-eight osteosarcoma patients (including 56 males, 63.6%) and a temporal independent validation cohort of 80 sarcoma patients (including 52 males, 65%).

3.0 T; T1-weighted spin-echo (SE), T2-weighted fast spin-echo (FSE), and contrast-enhanced T1-weighted (CE-T1WI) spin-echo sequences.

Tumors segmented on pre-treatment images were partitioned into four habitats using k-means clustering. Support Vector Machine (SVM) models were developed using features from habitats versus the entire tumor to predict 1-year recurrence. Unsupervised clustering identified prognostic subtypes.

The DeLong test was used to compare Area Under the Curve (AUC) values. Kaplan-Meier survival analysis (Log-rank test) and Chi-square tests were employed for prognostic stratification. A p-value < 0.05 was considered statistically significant.

The habitat-SVM model achieved the best performance, with an AUC of 0.839 (95% CI: 0.759-0.929) in the training cohort and 0.815 (95% CI: 0.782-0.999) in the temporal independent validation cohort. This performance was significantly superior to the best conventional model (AUC = 0.803). Unsupervised analysis identified four radiomic subtypes with significantly distinct recurrence rates (7.7%-76.7%) and disease-free survival outcomes.

MRI-based radiomic habitat analysis may help to characterize intratumoral heterogeneity in osteosarcoma, providing superior risk stratification for post-treatment recurrence. This non-invasive strategy offers a promising tool for individualized prognostic assessment. Limitations include the single-center design and small sample size.

3.

Stage 2.