Evaluation of the performance of both machine learning models using PET and CT radiomics for predicting recurrence...

Area of Science:

• Oncology
• Radiotherapy
• Medical Imaging

Background:

• Stereotactic body radiotherapy (SBRT) is a key treatment for non-small cell lung cancer (NSCLC).
• Predicting recurrence post-SBRT is crucial for personalized treatment planning.
• Radiomics, extracting quantitative features from medical images, shows promise in predicting cancer outcomes.

Purpose of the Study:

• To evaluate machine learning models for predicting NSCLC recurrence after SBRT.
• To compare the performance of models using computed tomography (CT) and positron emission tomography (PET) radiomic features.
• To identify optimal models for predicting local, regional, and distant metastasis.

Main Methods:

• Retrospective analysis of CT and PET images from 82 NSCLC patients treated with SBRT.
• Extraction and selection of 111 radiomic features from tumor delineations.
• Development and comparison of four machine learning models using selected features and AUC via 0.632+ bootstrap.

Main Results:

• Support vector machine with PET features achieved the highest performance for local recurrence (AUC=0.646).
• Naive Bayes with PET features showed the best performance for regional lymph node metastasis (AUC=0.611).
• Support vector machine with CT features was optimal for distant metastasis prediction (AUC=0.645).

Conclusions:

• Machine learning models utilizing radiomic features from CT and PET are effective for predicting NSCLC recurrence patterns post-SBRT.
• These models can inform individualized radiotherapy decisions and treatment strategies.
• Further development can enhance prediction accuracy and clinical utility.