Transitional zone prostate cancer: Performance of texture-based machine learning and image-based deep learning.

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Transitional-zone prostate cancer (TZPCa) detection is challenging, often obscured by benign prostatic hyperplasia (BPH).
• Accurate MRI-based detection of TZPCa is crucial for effective patient management.

Purpose of the Study:

• To compare the performance of texture-based machine learning and image-based deep learning for TZPCa detection.
• To evaluate these AI methods in differentiating TZPCa from BPH using MRI.
• To assess their potential as diagnostic or screening tools.

Main Methods:

• Texture analysis using machine learning (logistic regression, SVM, random forest) on T2WI MRI.
• Image-based deep learning (convolutional neural network) applied to TZPCa and BPH images.
• Validation using leave-one-out and 10-fold cross-validation, with performance metrics including AUC and ROC curves.

Main Results:

• Texture-based methods achieved high AUC (0.854-0.861) and specificity (0.710-0.775).
• Image-based deep learning demonstrated high sensitivity (0.946) with good AUC (0.802) but moderate specificity (0.643).

Conclusions:

• Texture-based machine learning shows promise as a supportive diagnostic tool for suspected TZPCa lesions.
• Image-based deep learning is effective as a screening tool for identifying suspicious TZPCa lesions due to its high sensitivity.