The Development of a Screening Model for Toxoplasmosis from Fundus Photography using Image Mining Techniques

Abstract

This research aimed to develop and evaluate the performance of screening models for toxoplasmosis using fundus photography. Data analysis was conducted following the Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology. Three image classification techniques were utilized: Decision Tree, Support Vector Machine (SVM), and k-Nearest Neighbors (k-NN). The hyperparameters of each technique were tuned to optimize model performance. The models were evaluated using Accuracy, F-measure, Sensitivity, and Specificity. Results revealed that the Decision Tree showed the best performance in image classification, with an Accuracy of 90.74%, F-measure of 83.15%, Sensitivity of 85.55%, and Specificity of 92.45%. Thus, it is suitable for developing a model for screening patients with toxoplasmosis from fundus images. The results of this study can support physicians' decision-making in diagnosing and screening patients with toxoplasmosis more accurately and predicting the risk of disease occurrence or complications.