الفهرس 
Deep learning for medical image diagnosis
التعلم العميق لتشخيص الصور الطبيةOnly 14 pages are availabe for public view
 |  | from | 83 |  |  |
| | | from | 83 | | |
Abstract
One of the most important tasks while developing medical diagnosis software system is diseases prediction. Artificial intelligence and neural networks are two main methods that have already been used to solve medical diagnosis problems. Deep Learning strategies have recently been popular in a variety of applications, including assisting in medical diagnosis. Patients can analyse disease based on clinical and laboratory symptoms with sufficient data and get a more efficient outcome for a particular disease in a very simple and timely manner. DL enhances the performance for medical image diagnosis by generating features directly from raw images. DL is a data-driven approach, it highly depends on the data used. Data limitation is always a critical problem when designing a DL model. This thesis provides a solution for medical image diagnosis with a limited number of medical images by proposing two different diagnosis models.The first one is a transfer learning-based model with a hinge loss function instead of the traditional softmax function. This diagnosis model for medical image classification utilizes the use of two different scenarios based on Inception V3 and Xception architectures.The second model utilizes the concept of ensemble learning by introducing an end-toend ensemble model. This proposed model is dependent on three pre-trained Convolutional Neural Network (CNN) (e.g. Xception, Inception, and VGG19 models). More layers were added to allow this model to discover the best concatenation weights among all three models. Both models are used for retinal diseases diagnosis and Alzheimer disease diagnosis