Fig 1.
Panel A. S. haematobium eggs in a urine sediment sample (10x ocular; 10x objective lens). Panel B. S. haematobium egg in a urine sediment sample (10x ocular; 40x objective lens).
Fig 2.
Representative scheme of sample visualization, image acquisition, image pre-processing, and image annotation procedures.
Sample/slide, image, annotated image, and label are represented. Schistosoma haematobium digital images were acquired with an integrated camera/smartphone camera in the Microbiology Laboratory of the Vall d’Hebron Drassanes International Health and Infectious Diseases Centre. Illustrations were obtained from open source resources.
Table 1.
Summary of the urine sediment sample image database.
Table 2.
Summary of Convolutional Neural Network training and performance parameters with the test image dataset.
DA: Data augmentation, mAP: mean average precision, YOLO: you only look once.
Fig 3.
Graphical representation of F-score/mAP0.5 of the different CNNs trained on the test dataset.
Orange dots represent the performance of CNNs trained with the 491-image dataset. Blue dots represent the performance of CNNs trained with the 1017-image dataset.
Fig 4.
Panel A. Digital image (400x) of a urine sediment sample with hematuria and leukocyturia and three Schistosoma haematobium eggs detected with the YOLOv5x trained model. Panel B. Digital image (100x) of a urine sediment sample with three S. haematobium eggs detected with the YOLOv5x trained model.
Table 3.
Summary of binary image classifier training and performance parameters on validation and test image datasets.