Prostate cancer prediction through a hybrid deep learning method applied to histopathological image

Area of Science:

• Medical image analysis
• Computational pathology
• Artificial intelligence in oncology

Background:

• Prostate cancer (PCa) diagnosis relies on the Gleason grading system using histopathological images.
• Manual Gleason grading is time-consuming and requires specialized expertise.
• Existing deep learning methods for Gleason grading face challenges with accuracy and computational cost.

Purpose of the Study:

• To develop an efficient and accurate deep learning network for automated Gleason grading of prostate cancer.
• To overcome the limitations of existing deep learning models in terms of accuracy and computational complexity.

Main Methods:

• A novel deep learning network integrating MobileNet, an Attention Mechanism (AM), and a capsule network was proposed.
• MobileNet was used for efficient feature extraction.
• The AM focused on relevant features, and the capsule network performed classification.
• The network was validated on the PANDA and Gleason-2019 datasets.

Main Results:

• The proposed network demonstrated high effectiveness in Gleason grading.
• Ablation studies confirmed the contribution of each component in the architecture.
• The network achieved 98.08% accuracy on the PANDA dataset and 97.07% on the Gleason-2019 dataset.

Conclusions:

• The developed deep learning network significantly outperforms existing approaches for prostate cancer Gleason grading.
• The integration of MobileNet, AM, and capsule networks offers a promising solution for accurate and efficient automated histopathological analysis in oncology.