MeFunX: A novel meta-learning-based deep learning architecture to detect fungal infection directly from microscopic images
Shubhankar Rawat, Bhanvi Bisht, Virender Bisht, Nitin Rawat, Aditya Rawat
Abstract
Fungal infections are a growing threat to human health. They can lead to a range of health problems and can be life-threatening. There are many impediments to the traditional diagnosis of fungal infections, such as a diminishing number of clinical mycologists, expensive procedures, high time consumption, and requirements for sensitivity and specificity. Therefore, early diagnosis of fungal infection is critical to effective treatment. In this paper, a novel meta-learning-based deep learning architecture, termed MeFunX, is proposed for the early detection of fungal infections from microscopic images. MeFunX architecture consisted of two convolutional neural network-based models as base learners and XGBoost as the meta-learner. To assess the proposed approach, standard metrics namely, accuracy, precision, recall and f1-score, were used. The fungal disease identification performance of MeFunX was compared with state-of-the-art architectures like VGG16, InceptionV3, ResNet, AlexNet, DenseNet, and EfficientNet. In addition, MeFunX was also benchmarked against its base learners and other meta-learning model with EfficientNet and ResNet as the base learners to demonstrate the effectiveness of the meta-learning architecture. Rigorous experimentation clearly signifies the superior performance of MeFunX, which achieved an overall accuracy of 92.49% for the early diagnoses of fungal infections in microscopic images.