Prognostic Prediction of Cancer Based on Radiomics Features of Diagnostic Imaging: The Performance of Machine...

Area of Science:

• Radiomics and Medical Imaging
• Machine Learning in Oncology
• Cancer Prognostics

Background:

• Tumor phenotypes are characterized by radiomics features from medical images.
• Accurate prediction of cancer prognosis is challenged by small sample sizes and data imbalance.

Purpose of the Study:

• To evaluate machine learning strategies for predicting cancer prognosis.
• To assess the performance of various algorithms using radiomics features.

Main Methods:

• Radiomics features were extracted from CT images of 422 non-small cell lung carcinoma (NSCLC) patients.
• Six machine learning methods (DT, BT, RF, SVM, GLM, DL-ANNs) were applied with 70:30 cross-validation.
• Performance was evaluated based on survival endpoints at 1, 3, 5, and 7 years.

Main Results:

• Random Forest (RF) achieved the highest AUC (0.938), followed by Boosted Trees (BT) (0.912).
• Traditional machine learning methods like RF and BT outperformed Deep Learning Artificial Neural Networks (DL-ANNs).
• DL-ANNs did not show a significant advantage over traditional methods for prognostic prediction.

Conclusions:

• Radiomics combined with machine learning, especially RF and BT, provides accurate cancer outcome prediction.
• These findings support radiomics as a valuable tool for informing cancer treatment strategies.
• Machine learning models offer a supportive reference for personalized cancer care planning.