دراسة مقارنة ونموذج هجين (CNN+SIFT) للكشف عن أنماط الأمراض الصدرية في الصور الطبية

Abstract

Deep learning has been highly successful in diagnosing numerous image-based disease models. Various deep learning applications used to diagnose diseases based on various types of medical images have become a hotspot of research in the fields of artificial intelligence and computer vision. Due to the rapid development of deep learning methods, diagnosing disease models based on digital images requires extremely high accuracy and timeliness. Furthermore, it is necessary to consider the inherent privacy and complexity of medical imaging.This research aims to test and evaluate deep learning models and classical pre-trained models. Advanced neural network models that have emerged in recent years, primarily convolutional neural network models(CNN) , were studied to detect three disease models in medical images: cancer, tuberculosis, and COVID-19. Each of these disease models has a specific distribution pattern within medical images. Studies have relied on analyzing multiple neural network models to improve the speed and accuracy of searching for one of the three disease models. The proposed study, in contrast, relies on an evaluation and development process to build a neural network model capable of detecting multiple disease patterns in medical images. The resulting model offers high accuracy and predictive speed.

The model has demonstrated good results in image classification. The model used in the AlixNET network succeeded with an accuracy of about 88%, in the Resnet network model it achieved an accuracy of 92.8%, in the U-net network model the accuracy was 89%, and in the proposed hybrid network model after using the SIFT algorithm the accuracy was 94%.