Development and Validation of a Machine Learning-Based Radiomics Model Using Ultrasound Image Features for Prostate...

Area of Science:

• Radiology
• Oncology
• Machine Learning

Background:

• Prostate cancer (PCa) risk stratification is crucial for effective treatment.
• Current methods rely on PSA levels, Gleason scores, and clinical T stage.
• Novel tools are needed for improved early diagnosis and decision-making.

Purpose of the Study:

• To develop a risk stratification model for PCa using ultrasound imaging and machine learning.
• To provide an effective tool for early PCa diagnosis and personalized treatment.

Main Methods:

• Retrospective analysis of 211 histopathologically confirmed PCa patients.
• Extraction of 135 quantitative radiomic features from ultrasound images.
• Application of machine learning algorithms (SVM, RF, LR) after feature selection.

Main Results:

• The Random Forest (RF) model demonstrated superior performance with high accuracy (90% training, 88% test) and AUC (0.87 training, 0.86 test).
• RF significantly outperformed SVM and Logistic Regression (LR) in AUC comparisons.
• Texture features from wavelet-transformed Gray Level Dependence Matrix (GLDM) were most predictive, emphasizing intratumoral heterogeneity.

Conclusions:

• An RF-based ultrasound radiomics model accurately stratifies PCa risk.
• The model shows remarkable performance in identifying high-risk prostate cancer patients.
• This approach offers a valuable tool for clinical decision-making in PCa management.