Detection of benign prostatic hyperplasia using RGB prostate images and deep learning

Area of Science:

• Digital Pathology
• Medical Image Analysis
• Artificial Intelligence in Healthcare

Background:

• Benign Prostatic Hyperplasia (BPH) is a common condition requiring accurate histopathological diagnosis.
• Deep learning models offer potential for automated analysis of complex medical images.
• Challenges include class imbalance and achieving high sensitivity for disease detection.

Purpose of the Study:

• To develop and evaluate a deep learning model for identifying BPH in RGB prostate histopathological images.
• To enhance image features using Adaptive CLAHE and integrate CNNs with BiLSTM and attention mechanisms.
• To address class imbalance and overfitting through adaptive focal loss and data augmentation.

Main Methods:

• Utilized a hybrid CNN-BiLSTM-Attention architecture for analyzing RGB prostate histopathology.
• Applied Adaptive CLAHE to the L-channel in the LAB color space for image enhancement.
• Implemented adaptive focal loss and image augmentation to manage data imbalance and prevent overfitting.

Main Results:

• The model achieved an AUC of 0.7220 on the validation set and 0.73 on the test set.
• Demonstrated high precision for normal tissue classification.
• Exhibited low recall for BPH detection, indicating a need for improved sensitivity.

Conclusions:

• The CNN-BiLSTM-Attention architecture shows promise as a diagnostic aid in digital pathology for BPH identification.
• Further improvements are needed to enhance the model's sensitivity in detecting BPH.
• Future work will focus on multi-class disease grading and improving BPH detection recall.