Advances and limitations of artificial intelligence-assisted identification of pathogenic fungi

ObjectivesWe developed and tested multiple computer-vision image classifiers, for their ability to identify a large set of common and rare pathogenic molds. Aim of the study was to create a comprehensive global benchmark towards the novel, emerging field of computer vision driven diagnostics of pathogenic microbes. If successfully implemented, a high-resource clinical setting could greatly benefit from this adjunct technique to supplement molecular sequencing and mass spectrometry driven methodologies, while in a low resource setting, it could provide enormous possibilities to enhance diagnostic precision in rural and remote geographies. MethodsWe selected 114 representative fungal pathogens represented by 123 strains obtained from the unique images implemented within the Atlas of Clinical Fungi, to serve as core dataset. The image classifiers were designed with a rigorous testing and evaluation strategy, at a yet unprecedented level of detail. We designed the framework, within the TENSORFLOW environment, testing multiple transfer-learning approaches, as well hybrid architectures comprising both, features of convolutional neural networks (CNN) and advanced vision transformers (ViT). ResultsWe achieved a global identification accuracy of > 88% for the validation partition with our best model (Test accu. 87%, Train. accu. 96%). Simulations indicated that extended training time would lead to further accuracy improvements, particularly with greater data richness. Our results also highlight complex de-black-boxing approaches in interpreting image classification, never shown for microbial computer vision diagnostics to date. DiscussionBesides quantitative limitations of representative strains per tested species, our approach reflects a significant scientific novelty to the field. While tested mainly on molds and a small subset of common bacteria as a control set, the methodology is universally applicable to yeasts and bacteria rendering the technique attractive for future diagnostics in the clinical setting.