Attention-Based Deep Convolutional Neural Network Versus Transfer Learning for Medical Image Classification and Disease Diagnosis / Maroua Azouz  

Public ISBD Titre : Attention-Based Deep Convolutional Neural Network Versus Transfer Learning for Medical Image Classification and Disease Diagnosis Type de document : texte imprimé Auteurs : Maroua Azouz, Auteur ; Nour Deghoul, Auteur ; Abdelouahab Moussaoui, Directeur de thèse Année de publication : 2022 Importance : 1 vol (94 f .) Format : 29cm Langues : Français (fre) Catégories : Thèses & Mémoires:Informatique Mots-clés : Transfer Learning Image Classification ViT Index. décimale : 004 Informatique Résumé : Cancer is a disease in which some of the body’s cells grow uncontrollably and spread to other parts of the body. recently the most two common type of diseases in our city Setif are Skin cancer and nodules in thyroid gland cancer. Skin Cancer Is the Cancer You Can See, this sentence is the the motto of Skin Cancer Foundation .When strange grains appear in the patient’s skin, he cannot differentiate whether it is cancer or just simple skin ulcers, so in most cases it is not detected except in advanced stages of the disease . Most patients with nodules in the thyroid gland are afraid to turning these nodes into cancerous masses, and with the spread of this disease in recent years, this question has become a concern for all patients. Recently, the world has become highly focused on deep learning and classification of medical images by building stable models in computer aided diagnosis [8], most often the model is using Convolutional neural networks. Now we propose a model based on the attention mechanism that based on the most important features in the image. The attention mechanism contributes to increasing the effectiveness of the model and achieving a better classification. The aim of this study is to improve the accuracy of a computer-aided diagnosis approach that medical professionals can easily use as an aid. In this thesis, we proposed the use of the transformer and transfer-learning mechanism to detect and classify skin cancer diseases and the type of thyroid nodules(benign or malignant). We collected features from three types of pre-trained models, EfficientNetB7, VGGNet16, Xception as feature extraction roles. Then as using the features as input to Vision Transformer (ViT) and then using neural networks for classification.As a result, the proposed approach achieved accuracy of 83.74% for skin dataset and 76.18% for thyroid dataset, and an other similar model using CNN model proposed by us is achieve 92.95% of accuracy for skin dataset and 88.18% of accuracy for thyroid dataset . We proposed, also the of use transfer learning for develop the pre trained model and applied the same datasets on ResNet50 and our new ResNet50,these models achieve 67.66%, 98.89% respectively for skin cancer data set and they achieve 67.03%, 90.97% respectively by the use of thyroid dataset. Côte titre : MAI/0587 En ligne : https://drive.google.com/file/d/1x9lI7fG4IlSYzf0xGs-E7AL9ynLyGu1h/view?usp=share [...] Format de la ressource électronique : pdf Attention-Based Deep Convolutional Neural Network Versus Transfer Learning for Medical Image Classification and Disease Diagnosis [texte imprimé] / Maroua Azouz, Auteur ; Nour Deghoul, Auteur ; Abdelouahab Moussaoui, Directeur de thèse . - 2022 . - 1 vol (94 f .) ; 29cm. Langues : Français (fre) Catégories : Thèses & Mémoires:Informatique Mots-clés : Transfer Learning Image Classification ViT Index. décimale : 004 Informatique Résumé : Cancer is a disease in which some of the body’s cells grow uncontrollably and spread to other parts of the body. recently the most two common type of diseases in our city Setif are Skin cancer and nodules in thyroid gland cancer. Skin Cancer Is the Cancer You Can See, this sentence is the the motto of Skin Cancer Foundation .When strange grains appear in the patient’s skin, he cannot differentiate whether it is cancer or just simple skin ulcers, so in most cases it is not detected except in advanced stages of the disease . Most patients with nodules in the thyroid gland are afraid to turning these nodes into cancerous masses, and with the spread of this disease in recent years, this question has become a concern for all patients. Recently, the world has become highly focused on deep learning and classification of medical images by building stable models in computer aided diagnosis [8], most often the model is using Convolutional neural networks. Now we propose a model based on the attention mechanism that based on the most important features in the image. The attention mechanism contributes to increasing the effectiveness of the model and achieving a better classification. The aim of this study is to improve the accuracy of a computer-aided diagnosis approach that medical professionals can easily use as an aid. In this thesis, we proposed the use of the transformer and transfer-learning mechanism to detect and classify skin cancer diseases and the type of thyroid nodules(benign or malignant). We collected features from three types of pre-trained models, EfficientNetB7, VGGNet16, Xception as feature extraction roles. Then as using the features as input to Vision Transformer (ViT) and then using neural networks for classification.As a result, the proposed approach achieved accuracy of 83.74% for skin dataset and 76.18% for thyroid dataset, and an other similar model using CNN model proposed by us is achieve 92.95% of accuracy for skin dataset and 88.18% of accuracy for thyroid dataset . We proposed, also the of use transfer learning for develop the pre trained model and applied the same datasets on ResNet50 and our new ResNet50,these models achieve 67.66%, 98.89% respectively for skin cancer data set and they achieve 67.03%, 90.97% respectively by the use of thyroid dataset. Côte titre : MAI/0587 En ligne : https://drive.google.com/file/d/1x9lI7fG4IlSYzf0xGs-E7AL9ynLyGu1h/view?usp=share [...] Format de la ressource électronique : pdf

Exemplaires (1)

Code-barres	Cote	Support	Localisation	Section	Disponibilité
MAI/0587	MAI/0587	Mémoire	Bibliothéque des sciences	Anglais	Disponible Disponible

---

1 (1 - 1 / 1)