Machine Learning for Laser Lesions
Abstract
Oceanit, using its RetinaView software and data collected by the U.S. Air Force developed and trained convolutional neural networks to detect and classify laser lesion injuries in three imaging modalities: Fundus imagery, hyperspectral imagery, and OCT imagery. Networks using the commonly collected Fundus and OCT imagery can detect lesions with 97% accuracy, providing a viable tool for non-expert clinicians with no experience in laser lesion injuries to detect them without additional imaging equipment. Further experiments show it is possible to predict the age of a given lesion, allow for the potential to tie an injury to a specific engagement or event.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 13, 2019
- Accession Number
- AD1081832
Entities
People
- Adam R. Boretsky
- Edward Pier
- Joel N. Bixler
- Zachary D. Stoecker-sylvia
Organizations
- Leidos