Master of Science
Epidemiology and Biostatistics
Cataract is the leading cause of blindness and vision loss globally. The implementation of artificial intelligence (AI) in the healthcare industry has been on the rise in the past few decades and machine learning (ML) classifiers have shown to be able to diagnose patients with cataracts. A systematic review and meta-analysis were conducted to assess the diagnostic accuracy of these ML classifiers for cataracts currently published in the literature. Retrieved from nine articles, the pooled sensitivity was 94.8% and the specificity was 96.0% for adult cataracts. Additionally, an economic analysis was conducted to explore the cost-effectiveness of implementing ML to diagnostic eye camps in rural Nepal compared to traditional diagnostic eye camps. There was a total of 22,805 patients included in the decision tree, and the ML-based eye camp was able to identify 31 additional cases of cataracts, and 2546 additional cases of non-cataract.
Summary for Lay Audience
Cataract is an eye disease that many older adults get. A cataract is a buildup of cloudiness in the human eye lens that can result in blurry and reduced vision. Fortunately, through early and proper screening procedures, cataracts can easily be detected, and cataract surgery can be performed to gain back vision. There has been an increasing use and implementation of artificial intelligence (AI) in the healthcare field and machine learning (ML) which is a subset of AI. In the field of ophthalmology, there are many developments for the use of ML classifiers that can automatically detect eye diseases (such as cataracts) by processing images of the eye through a computer algorithm.
In this thesis, a systematic review and meta-analysis were conducted to assess the diagnostic accuracy of current machine learning classifiers for cataracts in both published databases and unpublished literature. A total of 21 studies were included in the qualitative review, and a total of nine studies were included for the quantitative analysis. From the quantitative analysis, there was observed to be high diagnostic accuracy for identifying true cataract cases and true non-cataract cases, these values are known as sensitivity and specificity, respectively. For adult cataracts, there was a 94.8% sensitivity and 96.0% specificity.
Utilizing these results from the meta-analysis, a cost-effective analysis was conducted to test the economic feasibility of a ML cataract screening program to be implemented in a rural region. In Nepal, rural Nepalis may have access to temporary village-level primary eye care centres known as “diagnostic-screening and treatment camps (eye camps)”. The objective of this second study was to conduct a cost-effectiveness analysis of the theoretical implementation of a ML-based cataract screening eye camp in rural Nepal in order to assess if this new program is superior to the traditional eye camps. There was a total of 22,805 patients in each arm of the decision tree, and the ML-based eye camp could identify 31 additional cases of cataracts, and 2546 additional cases of non-cataract. This suggested that the ML-based eye camp was a more cost-effective method than the traditional eye camp in rural Nepal.
Cheung, Ronald, "Effectiveness of Machine Learning Classifiers for Cataract Screening" (2022). Electronic Thesis and Dissertation Repository. 8761.