AI in gastrointestinal disease detection: overcoming segmentation challenges with Coati optimization strategy

Today AI helps a lot within the healthcare industry in the handling and classification of diseases that affect people. Multiple AI-based computerized methodologies developed in recent years for diagnosing gastrointestinal diseases (GI) including polyps. Diagnosis of these diseases performed by human beings requires much more time is expensive and requires expertise in this field. Thus the comprised methodologies that help the doctors as a second opinion is greatly requested. The prime issues faced by these technologies are segmentation and finding the precise regions where the infected area is there since the affected region has varying locations and shapes. This is due to the segmentation inaccuracy leading the negative impacts on feature learning and further affecting the preciseness in classification. Thus this paper introduced a novel automatic strategy for diagnosing GI disease which is designed with some of the following prime phases, from preprocessing to classification. The input GI images are gained from the Kvasir Dataset contains multiple classes like polyps, ulcerative colitis, etc., and the CVC-ClinicDB dataset which includes a single class of polyps images. For extracting the features a Coati optimization algorithm is employed in the feature learning model which helps in finding and focusing on the relevant features or information and makes the model more robust. A hybrid U-net model is employed for the pixel-level segmenting process, which can minimize the errors and enhance the segmentation process. Along with the Mask region based convolutional neural network (Mask RCNN) is employed to classify the segmented images, which enhances the accuracy of diagnosis and also the affected area's localization. Experimental outputs concluded that the proposed method offered realistic and pragmatic solutions for diagnosing GI diseases with enhanced accuracy of 98.8%, 98.2% recall, 97% F1-score, and 97.5% preciseness value. Further, it improved the prognosis pace by curtailing the training time of the model to 15 s.