DOI: 10.4103/jmp.jmp_32_26 ISSN: 0971-6203

Magnetic Resonance Imaging-based Prostate Cancer Diagnosis Using Principal Component Analysis and Machine Learning Algorithms

N. Samta, C. S. Sureka, Sivananthan Sarasanandarajah, R. K. Jeevanram
Show PDF Cite

Purpose:

Prostate cancer is a leading cause of cancer-related mortality among men, necessitating accurate and early diagnostic methods. This study evaluates the effectiveness of principal component analysis (PCA) combined with machine learning (ML) for classifying benign and malignant prostate lesions using magnetic resonance imaging (MRI

). Materials and Methods:

MRI data from 26 biopsy-confirmed prostate cancer patients were obtained from The Cancer Imaging Archive. Preprocessing included resizing and intensity normalization of dynamic contrast-enhanced MRI images. PCA was applied to extract dominant features capturing key tissue variations. A comparative analysis was performed using three feature representation strategies: raw pixel-based features, PCA-transformed features, and radiomics-based features. The extracted features were classified using Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) classifiers.

Results:

PCA-based features demonstrated separation between benign and malignant tissue regions. Statistical analysis using an independent two-sample t -test (Welch’s correction) revealed a highly significant difference in mean PC1 values between the two groups ( P < 2.2 × 10 ⁻16 ). Among the classifiers, RF achieved the highest performance, followed by SVM and KNN, as confirmed by receiver operating characteristic analysis (area under the curve [AUC] = 0.909, 0.805, and 0.785, respectively), where higher AUC values indicate the model’s ability to distinguish between benign and malignant lesions. Comparative evaluation indicated that PCA-based models outperformed raw pixel-based approaches, while radiomics-based features yielded additional gains.

Conclusion:

PCA-based feature extraction may enhance classification performance by reducing dimensionality and noise in MRI data. The proposed PCA–ML framework offers an interpretable, efficient, and clinically relevant approach for prostate cancer diagnosis.

More from our Archive