PET image nonuniformity texture features for metastasis risk prediction in osteosarcoma

Area of Science:

• Oncology
• Radiology
• Medical Imaging

Background:

• Osteosarcoma metastasis risk and chemotherapy response are linked to tumor heterogeneity.
• Ki-67 expression is a known predictor of metastasis in osteosarcoma.
• Integrating imaging and molecular data can enhance predictive model accuracy.

Purpose of the Study:

• To evaluate the accuracy of combining 18F-fluorodeoxyglucose PET image texture features and Ki-67 expression for predicting metastasis in osteosarcoma.
• To assess the predictive value of gray-level run length matrix (GLRLM) run length nonuniformity (RLNU) for neoadjuvant chemotherapy response (NACR) and metastatic events.

Main Methods:

• Collected PET images and clinical data from 82 osteosarcoma patients.
• Extracted quantitative texture features from pre-treatment PET images.
• Utilized machine learning (random forest) with GLRLM RLNU, Ki-67, and NACR for metastasis prediction.

Main Results:

• GLRLM RLNU was significantly associated with NACR and metastatic risk (P < 0.05).
• A random forest model incorporating GLRLM RLNU, Ki-67, and NACR achieved 0.91 accuracy in predicting metastatic risk.
• PET image texture nonuniformity demonstrated high accuracy in predicting NACR and metastatic risk.

Conclusions:

• Combining PET image texture nonuniformity with Ki-67 expression and clinical data improves metastasis prediction accuracy in osteosarcoma.
• This multimodal approach offers a promising tool for metastasis risk stratification.
• Personalized treatment strategies for osteosarcoma can be supported by this enhanced predictive model.