RESULT: We tested Naive Bayes, Logistic Regression, KNN, J48, Random Forest, SVM, and Deep Neural Network algorithms to ASD screening dataset and compared the classifiers' based on significant parameters; sensitivity, specificity, accuracy, receiver operating characteristic, area under the curve, and runtime, in predicting ASD occurrences. We also found that most of previous studies focused on classifying health-related dataset while ignoring the missing values which may contribute to significant impacts to the classification result which in turn may impact the life of the patients. Thus, we addressed the missing values by implementing imputation method where they are replaced with the mean of the available records found in the dataset.

METHODS: 120 primary pterygium participants were selected from patients who visited an ophthalmology clinic. We adopted image analysis software in calculating the size of invading pterygium to the cornea. The marking of the calculated area was done manually, and the total area size was measured in pixel. The computed area is defined as the area from the apex of pterygium to the limbal-corneal border. Then, from the pixel, it was transformed into a percentage (%), which represents the CPTA relative to the entire corneal surface area. Intra- and inter-observer reliability testing were performed by repeating the tracing process twice with a different sequence of images at least one (1) month apart. Intraclass correlation (ICC) and scatter plot were used to describe the reliability of measurement.

RESULTS: The overall mean (N=120) of CPTA was 45.26±13.51% (CI: 42.38-48.36). Reliability for region of interest (ROI) demarcation of CPTA were excellent with intra and inter-agreement of 0.995 (95% CI, 0.994-0.998; P<0.001) and 0.994 (95% CI, 0.992-0.997; P<0.001) respectively. The new method was positively associated with corneal astigmatism (P<0.01). This method was able to predict 37% of the variance in CA compared to 21% using standard method.