Detecting and Classifying Mycetoma in Histopathological Images Using DenseNet and U-Net.

Mycetoma, recognised by the WHO as a Neglected Tropical Disease, has significant diagnostic hurdles which lead to severe health consequences. Mycetoma can be caused by certain types of bacteria (actinomycetoma) or fungi (eumycetoma). Identifying whether mycetoma is bacterial or fungal plays a significant role in treatment planning, and an incorrect diagnosis can seriously affect patient prognosis and outcome. This paper explored the feasibility of using state-of-the-art artificial intelligence (AI) computer vision methods to effectively detect and diagnose mycetoma grains from histopathological microscopic images. This study utilised a dataset of 863 histopathological images from the Mycetoma Research Center in Khartoum, Sudan, provided by the MICCAI mAIcetoma challenge (https://mycetoma.edu.sd/?p=5242). We adapted U-Net segmentation and DenseNet classification models for the localisation and type prediction of mycetoma grains. Our results demonstrate high performance with classification accuracy of 94.52% and Dice Score of 0.8362. Code is available at: https://github.com/oliverjm1/mycetoma_segmentation.