<![CDATA[Gastrointestinal diseases are health problems that require the help of medical image analysis to improve accuracy and consistency in clinical decision-making. The main challenges with multilabel classification are visual complexity and morphological similarity. The aim of this study was to develop and evaluate an in-depth learning approach to build the first gastrointestinal tract image-based Clinical Decision Support System (CDSS). This dataset is publicly available and consists of 14 classes of gastrointestinal conditions with a total of 8,750 images, including 7,000 training images and 1,750 test images with a balanced distribution of classes. Four pre-trained convolutional neural network architectures were compared, namely MobileNetV2, MobileNetV3-Small, EfficientNet-B0, and ResNet50. The evaluation metrics used were accuracy, precision, recall, F1-score, confusion matrix, and case study inference. The experimental results showed that ResNet50 outperformed the others with 88.97% accuracy, 89.13% accuracy, 88.97% recall, and 88.94% F1-score, with multiple class analyses. Single-case inference testing on six randomly selected test images obtained a confidence value between 90-99%. The selected model is integrated into the mobile CDSS app to provide a level of confidence along with the predicted outcome. This method will likely allow for fundamental image-based evaluation to be more consistent and accountable in supporting clinical decision-making.]]>
Copyrights © 2026
