The Role of Radiomic Analysis and Different Machine Learning Models in Prostate Cancer Diagnosis.

Area of Science:

• Medical Imaging
• Oncology
• Artificial Intelligence

Background:

• Prostate cancer (PCa) is the most common cancer in men, necessitating precise grading for effective treatment.
• Biparametric Magnetic Resonance Imaging (bpMRI) radiomics combined with machine learning (ML) shows potential for enhancing PCa diagnosis and prognosis.

Purpose of the Study:

• To evaluate the diagnostic efficiency of seven ML models in differentiating PCa grades.
• To assess the impact of input variables, including radiomic features from T2-weighted (T2W) and diffusion-weighted (DWI) MRI and Prostate-Specific Antigen (PSA) values, on model performance.

Main Methods:

• A total of 214 men undergoing bpMRI were included.
• Radiomic features were extracted from T2W and DWI sequences.
• Seven ML algorithms were trained and tested using these features, with and without the inclusion of PSA index, and evaluated via receiver operating characteristic curve analysis.

Main Results:

• ML models using only T2W and DWI radiomic features showed limited clinical utility (AUC 0.703–0.807).
• Incorporating the PSA index significantly improved model performance across all grades and locations (AUC 0.784–1.00).
• Model performance was highly dependent on the input parameters used.

Conclusions:

• ML combined with bpMRI radiomics can aid in solving differential diagnostic problems in prostate cancer.
• The inclusion of the PSA index is critical for optimizing ML model efficiency in PCa grading.
• Further optimization of analysis methods is essential for clinical application.