Prostate Cancer Detection from MRI Using Efficient Feature Extraction with Transfer Learning.

Area of Science:

• Oncology
• Computer Science
• Medical Imaging

Background:

• Prostate cancer is a prevalent global health concern requiring prompt and accurate diagnosis for effective treatment.
• Machine learning (ML) offers promising avenues for enhancing diagnostic precision in oncology.
• Deep learning (DL) models show potential in extracting complex features from medical imaging data.

Purpose of the Study:

• To investigate the efficacy of various deep learning models (VGG16, VGG19, ResNet50, ResNet50V2) for feature extraction in prostate cancer diagnosis.
• To evaluate the performance of a random forest classifier in conjunction with DL-extracted features for prostate cancer classification.
• To address dataset limitations using transfer learning for improved model generalization.

Main Methods:

• Comparative analysis of VGG16, VGG19, ResNet50, and ResNet50V2 for feature extraction from prostate cancer images.
• Application of a random forest classifier to classify features extracted by DL models.
• Utilizing transfer learning techniques to train DL models on limited annotated prostate cancer datasets.

Main Results:

• ResNet50 achieved the highest accuracy (99.64%) in extracting significant features from prostate cancer images.
• The combination of DL feature extraction and random forest classification demonstrated high efficacy in prostate cancer detection.
• Transfer learning enhanced the generalization capability of DL models with limited data.

Conclusions:

• Deep learning models, particularly ResNet50, are highly effective for feature extraction in prostate cancer diagnosis.
• The integration of DL feature extraction with random forest classification provides a robust framework for accurate cancer identification.
• This research supports the development of reliable, interpretable ML-based diagnostic tools for early and precise prostate cancer detection.