Fast Multiphoton Microscopic Imaging Joint Image Super-Resolution for Automated Gleason Grading of Prostate Cancers.

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Oncology

Background:

• The Gleason grading system is crucial for prostate cancer quantification.
• Multiphoton microscopy (MPM) offers high-resolution imaging but faces challenges with speed and quality.
• Automated Gleason grading requires efficient and accurate imaging techniques.

Purpose of the Study:

• To introduce a fast deep learning-based multiphoton microscopy (MPM) imaging method for automated Gleason grading.
• To enhance MPM imaging speed and quality using deep learning for improved prostate cancer diagnosis.

Main Methods:

• A deep learning architecture (SwinIR) was employed for image super-resolution to improve low-resolution MPM images.
• Image acquisition speed was increased from 7.55s/frame to 0.24s/frame.
• A Swin Transformer classification network was utilized for automated Gleason grading.

Main Results:

• Super-resolution imaging improved the quality of low-resolution MPM images.
• Classification accuracy and Macro-F1 scores were 90.9% for high-resolution images and 89.9% for super-resolution images.
• Super-resolution images demonstrated comparable performance to high-resolution images for Gleason grading.

Conclusions:

• MPM combined with deep learning-based super-resolution and automated classification shows potential for real-time prostate cancer diagnosis.
• This approach can significantly accelerate the diagnostic process for prostate cancer.
• The method offers a viable pathway towards a real-time clinical diagnostic tool.