A striking histological overlap between diverse but related disorders makes illness diagnosis difficult. There is a significant clinical need to create computational algorithms that will allow doctors to convert diverse biological images into accurate and quantitative diagnosis. This requirement is especially pressing in the case of small bowel enteropathies, environmental enteropathy (EE), and celiac disease (CD). The study expanded on preliminary findings by creating an AI-based image analysis platform for these enteropathies that uses deep learning convolutional neural networks (CNNs). CNNs, one with multizoom architecture, have been created for the EE and CD image analysis platform. In order to depict the decision-making process for model classification of each disease, gradient Weighted Class activation mappings have been employed.
461 high-resolution biopsy pictures were obtained from 150 children. Median age was around the same sex distribution for 37.5months; 77 males. ResNet50 and shallow CNN showed an accuracy of 98% and 96% for case detection, which rose to 98,3% in an ensemble. Grad-CAMs showed that models can learn many morphological microscopic aspects for EE, CD and inspection. In order to detect biologically important microscopic features and simulate human pathology decision-making processes, the IA-based platform has shown good classification accuracy for small intestinal enteropathies.
