Glaucoma Functional Severity Staging System Based on Artificial Intelligence
For a study, researchers sought to develop an accurate and user-friendly glaucoma staging method based on visual fields (VFs). The models were developed and validated using a total of 13,231 VFs from 8,077 patients and 8,024 VFs from 4,445 subjects, respectively. Using an unsupervised machine learning approach, they could identify clusters with equivalent VF values. They labeled each cluster according to its unique mean deviation (MD). The investigators identified the appropriate MD thresholds for accurately differentiating clusters based on the Bayes minimum error principle. They evaluated the validity of the staging system and validated their results using a different validation dataset. About 4 clusters were found using the unsupervised k-means technique; they had, respectively, 6,784, 4,034, 1,541, and 872 VFs with average MDs of 0.0 dB (±1.4: SD), −4.8 dB (±1.9), −12.2 dB (±2.9), and −23.0 dB (±3.8). The optimal MD thresholds for discriminating between healthy eyes and those in the early, intermediate, and advanced stages of glaucoma were found to be -2.2, -8.0, and -17.3 dB using the supervised Bayes minimal error classifier. The accuracy of the glaucoma staging system, based on the MD thresholds in relation to the initial k-means clusters, was 94%. Based on unsupervised and supervised machine learning, study group discovered that 4 severity levels based on MD thresholds of 2.2, 8.0, and 17.3 dB are the best number of severity stages. This glaucoma staging approach was objective, precise, simple, and consistent, making it the perfect choice for use in both routine clinical practice and glaucoma research.