Abstract
Cryptococcus neoformans is an opportunistic fungal pathogen with significant medical importance, especially in immunosuppressed patients. It is the causative agent of cryptococcosis. An estimated 220,000 annual cases of cryptococcal meningitis (CM) occur among people with HIV/AIDS globally, resulting in nearly 181,000 deaths. The gold standards for the diagnosis are either direct microscopic identification or fungal cultures. However, these diagnostic methods need special types of equipment and clinical expertise, and relatively low sensitivities have also been reported. This study aims to produce and implement a deep-learning approach to detect C. neoformans in patient samples. Therefore, we adopted the state-of-the-art VGG16 model, which determines the output information from a single image. Images that contain C. neoformans are designated positive, while others are designated negative throughout this section. Model training, validation, testing, and evaluation were conducted using frameworks and libraries. The state-of-the-art VGG16 model produced an accuracy and loss of 86.88% and 0.36203, respectively. Results prove that the deep learning framework VGG16 can be helpful as an alternative diagnostic method for the rapid and accurate identification of the C. neoformans, leading to early diagnosis and subsequent treatment. Further studies should include more and higher quality images to eliminate the limitations of the adopted deep learning model.