A deep learning network for Gleason grading of prostate biopsies using EfficientNet

Area of Science:

• Oncology
• Computer Science
• Medical Imaging

Background:

• Accurate cancer severity grading is vital for diagnosis and treatment planning.
• Gleason's score is a standard grading system for prostate cancer.
• Manual grading of prostate cancer microscopic images is time-consuming and labor-intensive.

Purpose of the Study:

• To develop an automated system for prostate cancer Gleason grading using deep learning.
• To improve the efficiency and accuracy of prostate cancer diagnosis.

Main Methods:

• A novel deep learning network based on the EfficientNet architecture was proposed.
• The network incorporates a compound scaling method for dimensional balancing.
• An additional attention branch was integrated into EfficientNet-B7 for precise feature weighting.

Main Results:

• The proposed model integrates an attention branch with EfficientNet for Gleason grading.
• Training was performed on H&E-stained prostate cancer tissue samples from the Harvard Dataverse dataset.
• The model achieved a Kappa score of 0.5775, outperforming existing methods.

Conclusions:

• The developed deep learning network effectively automates prostate cancer Gleason grading.
• The integration of an attention branch enhances the model's performance.
• This approach offers a more efficient and accurate alternative to manual grading.