APPLICATION OF TRANSFER LEARNING FOR THE DETECTION OF TUBERCULOSIS FROM MICROSCOPIC SLIDE AND CHEST X-RAY IMAGES / MISHABU KORAMAR BOKO LAWAL; SUPERVISOR: Asst. Prof. Dr. Kamil YURTKAN
Dil: İngilizce 2022Tanım: 46 sheets; 31 cm. Includes CDİçerik türü:- text
- unmediated
- volume
Materyal türü | Geçerli Kütüphane | Koleksiyon | Yer Numarası | Durum | Notlar | İade tarihi | Barkod | Materyal Ayırtmaları | |
---|---|---|---|---|---|---|---|---|---|
Thesis | CIU LIBRARY Tez Koleksiyonu | Tez Koleksiyonu | YL 2455 L29 2022 (Rafa gözat(Aşağıda açılır)) | Kullanılabilir | Bioengineering Department | T2768 | |||
Suppl. CD | CIU LIBRARY Görsel İşitsel | YL 2455 L29 2022 (Rafa gözat(Aşağıda açılır)) | Kullanılabilir | Bioengineering Department | CDT2768 |
CIU LIBRARY raflarına göz atılıyor, Raftaki konumu: Tez Koleksiyonu, Koleksiyon: Tez Koleksiyonu Raf tarayıcısını kapatın(Raf tarayıcısını kapatır)
Thesis (MSc) - Cyprus International University. Institute of Graduate Studies and Research Bioengineering Department
Includes bibliography (sheets 37-40)
ABSTRACT
Tuberculosis can be detected from microscopic slides and chest X-ray images. The motivation of the automatic detection systems for tuberculosis is the high cases of tuberculosis. The particular detection problems can be solved by employing an AI-driven model which learn features. In this thesis, an automated detection model is presented using X-ray and microscopic slide images to detect tuberculosis into positive and negative cases. A pretrained Alexnet model is employed. The experiments are done on Chest X-ray dataset made available on Kaggle repository and microscopic slide images from both Federal Medical Center Zamfara and Kaggle repository. For classification of tuberculosis using microscopic slide images, the model recorded 90.56 testing accuracy, 97.78% testing sensitivity and 83.33% testing specificity for 70: 30 training and testing splits, 100% testing accuracy, 100% testing sensitivity and 100% testing specificity for 80:20 split, 96.67% testing accuracy, 93.33% testing sensitivity and 100% testing specificity for 90:10 training and testing split and 94.66% average testing accuracy, 98.33% average testing sensitivity and 91.00 % average testing specificity after cross validation. For classification of tuberculosis using X-ray images, the used method achieved 93.89% testing, 96.67% testing sensitivity and 91.11% testing specificity for 70:30 training and testing splits, 94.17% testing accuracy, 98.33% testing sensitivity and 90% testing specificity for 80:20 training and testing split, 98.33% testing accuracy, 100% testing sensitivity and 96.67% testing specificity for 90:10 training and testing split and 94.00% average testing accuracy, 98.33% average testing sensitivity and 98.33% average testing specificity after cross validation. Our results are following the notion that pretrained convolutional neural network models can take part for categorizing medical images with acceptable accuracy. These particular models can be used as a confirmatory tool for diagnosis of both pneumonia and tuberculosis, augmenting miss diagnosis and bring an option for the higher and monotonous workload faced by specialists, radiologists and pathologists in Federal Medical Centre Zamfara.
Keywords: Chest X-ray, deep learning, microscopic slide, pretrained network, AlexNet, Tuberculosis, transfer learning.