Identification and Localization of Indolent and Aggressive Prostate Cancers Using Multilevel Bi-LSTM

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate identification of indolent and aggressive prostate cancers is crucial for effective treatment, but current methods face challenges with limited accuracy and distinguishing cancerous from healthy tissue.
• Existing prostate cancer detection techniques often suffer from false positives and negatives due to reliance on imprecise ground truth labels and overlooking pathological characteristics.

Purpose of the Study:

• To develop a comprehensive framework for accurate identification and localization of prostate cancers, regardless of aggressiveness.
• To overcome the limitations of current diagnostic approaches by integrating advanced deep learning techniques.

Main Methods:

• Utilized a multilevel bidirectional long short-term memory (Bi-LSTM) model for prostate cancer identification and localization.
• Employed U-Net segmentation with ResNet-101 and a channel-based attention module for image pre-processing and feature map generation.
• Integrated statistical, global hybrid, and ResNet-101 feature maps, further optimized with channel and spatial attention mechanisms for enhanced detection.

Main Results:

• The framework achieved high performance metrics: 96.72% accuracy, 96.17% sensitivity, and 96.17% specificity on dataset 1.
• On dataset 2, the model demonstrated strong results with 94.41% accuracy, 93.10% sensitivity, and 94.96% specificity.
• The proposed method significantly surpassed the efficiency of alternative existing methods in prostate cancer detection.

Conclusions:

• The developed framework offers a promising approach for improving the diagnosis and localization of both indolent and aggressive prostate cancers.
• The integration of multilevel Bi-LSTM, U-Net segmentation, and attention mechanisms provides robust and accurate cancer identification.
• The model's validated effectiveness on distinct datasets highlights its potential for clinical application in prostate cancer management.