An Approach to Classify Ocular Toxoplasmosis Images using Deep Learning Models

Ocular Toxoplasmosis (OT) is a very well-known eye inflammation that is occurred from a parasite named Toxoplasma gondii (T. gondii) that can lead to several vision problems. Diagnosis traditionally involves clinical examinations and imaging, which can be complex and expensive, requiring specialized expertise. Recently, the interest of using the domain of artificial intelligence (AI) for diagnosing various eye diseases by analyzing fundus images has been increasing. To utilize this field, in this paper, a pre-trained model named VGG16 is built using deep learning (DL) algorithms. Basically, this is trying to use Artificial Neural Network (ANN) and Convolutional Neural(CNN) to classify the fundus images that have three categorical forms of OT. The dataset was collected with 355 images, and after augmentation, the data increased to around 700 images. Augmentation is done after rotating, flipping, shifting, shearing, and zooming. Data is also preprocessed to make the visions clear for the models to predict more accurately. Later, the performance of this proposed model is measured by some evaluating factors, such as the F1-score, precision, recall, loss, accuracy, and confusion matrix. In terms of overall performance among ANN and CNN models, ANN performed better with an accuracy of 98%, whereas to get acceptable accuracy, the CNN model required changing the images to grayscale, resulting in 95% accuracy.