NEWS & OPINION 42 EWAP MARCH 2023 by Tirth J. Shah, MD,* Zachary Q. Mortensen, MD,* Michael D. Abramoff, MD, and Thomas A. Oetting, MD *Co-first authors Contact information Mortensen: zachary-mortensen@uiowa.edu Oetting: thomas-oetting@uiowa.edu Shah: tirth-shah@uiowa.edu Review of ‘LOCSIII-based artificial intelligence program for automatic cataract grading’ T he rapid rise of artificial intelligence (AI) has revolutionized the field of medicine. As an image-centric specialty, ophthalmology stands at the frontier of AI applications. The implementation of AI has already provided large-scale early screening and detection of eye diseases, improved health disparities through better access, and decreased cost, in both the U.S. and low-income countries. Comparatively, the development of AI for evaluation of the lens is still in its infancy. 1,2,3 Lu et al. set out to establish and validate an AI-assisted automatic cataract grading program based on the Lens Opacities Classification System III (LOCSIII). 4 Methods The authors prospectively analyzed an internal dataset of slit lamp photographs of the anterior segment of cataract-affected eyes taken between 2018 and 2020 at the Fudan University Department of Ophthalmology. The group also obtained an external dataset of slit lamp photographs from March 2018 to August 2019 from the Pujiang Eye Study in Shanghai. All photographs were taken under pharmacologic mydriasis. The internal dataset was used to train, validate, and test, while the external dataset was used for testing. Each patient received slit beam imaging for nuclear cataract, diffuse illumination imaging for cortical cataracts, and retroillumination for posterior subcapsular cataracts. Eyes with small pupils, blurred region of interest, or any corneal disease that interfered with lens observation were excluded. The images were graded based on LOCSIII. Nuclear cataracts were graded on a scale from 1.0 to 6.0 with respect to nuclear color. Cortical and posterior subcapsular cataracts were graded on a scale from 1.0 to 5.0. The images were graded by at least two experienced ophthalmologists, and the reference standard was the average of the two experts. A grade of 3.0 or greater was regarded as “moderate to severe” requiring referral. Advanced deep learning algorithms, including Faster R-CNN and ResNet, were applied to identify the capture modes, annotate regions of interest, and grade the cataracts. Interobserver repeatability among the various ophthalmologists was assessed with intraclass correlation This article originally appeared in the December 2022 issue of EyeWorld. It has been slightly modified and appears here with permission from the ASCRS Ophthalmic Services Corp. University of Iowa residents and faculty; top row: Arnulfo Reyes, MD, Brandon Baksh, MD, Patrick Donegan, MD, Cheryl Wang, MD, Joanna Silverman, MD, Chad Lewis, MD, Samuel Tadros, MD, Bilal Ahmed, MD, Matthew Meyer, MD; middle row: Caroline Yu, MD, Sean Rodriguez, MD, Mahsaw Motlagh, MD, Aaron Dotson, MD; bottom row: Program Director Pavlina Kemp, MD, Tirth Shah, MD, Michael Abramoff, MD, Zachary Mortensen, MD. Source: University of Iowa
RkJQdWJsaXNoZXIy Njk2NTg0