[Deep Learning in Pathology: Applications and Challenges in Ophthalmic Pathology]

Area of Science:

• Ophthalmology
• Pathology
• Computer Science

Background:

• Deep learning (DL) excels in image analysis, presenting opportunities for histopathological applications.
• This study explores DL's potential utility within ophthalmic pathology, a specialized field.
• The research also assesses the volume of relevant cases and technical requirements.

Purpose of the Study:

• To identify potential deep learning (DL) applications relevant to ophthalmic pathology.
• To estimate the number of suitable cases and the technical effort required for DL implementation.

Main Methods:

• A literature review identified DL models applicable to general pathology.
• The number of suitable ophthalmopathology cases from 2019 was quantified.
• Data volume and scanning time were estimated for potential DL models.

Main Results:

• Seven areas of interest for DL in ophthalmic pathology were identified, including ROI determination, image classification, tumor fraction mapping, inflammation differentiation, cutaneous tumor classification, lymphoma classification, and outcome prediction.
• 831 cases (43%) were suitable for DL models within a year.
• Generating whole slide images (WSI) for these cases would require 630 GB of storage and 35 hours of scanning.

Conclusions:

• Several DL approaches are relevant to ophthalmic pathology, though most require specialized training.
• Implementing DL necessitates a robust IT infrastructure for creating and storing WSI, which is technically feasible.
• Future research should concentrate on the practical application of DL in ophthalmic pathology.