Faculty of Computing ,Library and Information Science (FCLIS).

Permanent URI for this collection

http://hdl.handle.net/20.500.12493/8

Browse

Browsing Faculty of Computing ,Library and Information Science (FCLIS). by Subject "Deep Learning"

Now showing 1 - 3 of 3
  • Thumbnail Image
    ItemOpen Access
    Comparison of Deep Learning Techniques in Detection of Sickle Cell Disease.
    (Kabale University, 2023) Mabirizi, Vicent; Kawuma, Simon; Kyarisiima, Addah; Bamutura, David; Atwiine, Barnabas; Nanjebe, Deborah; Oyesigye, Adolf Mukama
    Recently, the transfer learning technique has proved to be powerful in enhancing the development of deep learning methods for sickle cell disease (SCD) detection as a complement to the clinical method where a hemoglobin electrophoresis machine is used. This is evidenced by some models and algorithms with ≥90% prediction accuracy. From the literature, most of the proposed methods are trained and tested on pre-trained deep learning models like VGG16, VGG19, ResNet, Inception_V3, and ReNet. However, training and testing of these methods are limited to one model and separate datasets which may lead to biased results due to implementation in a variation of these models which affects the results produced. To this end, there exists a need to evaluate the SCD models using the same dataset. Thus, in this research study, we carried out a comparative investigation and evaluated predominate pre-trained models used to detect SCD using the same dataset to ascertain which one has the best accuracy. We used a secondary dataset obtained from an online dataset. In our study, we have discovered that Inception V3 yielded the highest accuracy of 97.3% followed by VGG19 at 97.0%, VGG16 at 91%, ResNet50 at 82% and ReNet at 67%, and the CNN-scratch model achieved 81% accuracy. Results from our study will aid researchers and industry practitioners in making decisions on the best deep-learning model to use while detecting SCD.
  • Thumbnail Image
    ItemOpen Access
    Diagnosis and Classification of Tuberculosis Chest X-ray Images of Children Less Than 15 years at Mbarara Regional Referral Hospital Using Deep Learning.
    (Kabale University, 2024) Kawuma, Simon; Kumbakumba, Elias; Mabirizi, Vicent; Nanjebe, Deborah; Mworozi, Kenneth; Mukama, Adolf Oyesigye; Kyasimire, Lydia
    Tuberculosis (TB) is an underestimated cause of death in children, with only 45% of cases correctly diagnosed and reported. It is estimated that 1.12 million TB cases occurred among newborns, children, and adolescents aged less or equal 14 years. In Uganda, TB prevalence is 8.5% in children and 16.7% in adolescents. Treatment and diagnosing TB is challenging and its high mortality rate is due to many lacks in the diagnosis of this illness especially among children. As a strategy to curb TB mortality rate in children, there exists a need to improve and expedite the screening for TB among children. Chest X-ray (CXR) is commonly used in TB burdened countries like Uganda to diagnose TB patients but interpretation of the patient’s radiograph needs skilled radiologists who are few. To this end, this research aims to close the TB mortality gap in children by applying AI, primarily deep learning techniques, to detect TB in children. The study created five models, one from scratch and four pre-trained Transfer Learning (TL) and were trained and verified using digital CXR radiograph images of children who visit the TB clinic at Mbarara Regional Referral Hospital. The model classifies clinical images of patients into normal or Tuberculosis. TL models; VGG16, VGG19, Inception V3, and ResNet50 outperformed scratch model with validation accuracy of 79.91%, 69.21%, 53.0%, 51.09% and 50.01% respectively. We hope that once the deep learning models are implemented and adopted by the radiologist, it will reduce the time spent by radiologist while analysing CXR images.
  • Thumbnail Image
    ItemOpen Access
    Masked and unmasked Face Recognition Model Using Deep Learning Techniques. A case of Black Race.
    (Kabale University, 2023) Mabiriz,I, Vicent; Ampaire, Ray Brooks; Muhoza, B. Gloria
    Currently, many institutions of higher learning in Uganda are faced with major security threats ranging from burglary to cyber threats. Consequently, the institutions have recruited and deployed several trained personnel to offer the desired security. As human beings, these personnel can make errors either by commission or omission. To overcome the limitation of trained security personnel, many face recognition models that detect masked and unmasked faces automatically to allow access to sensitive premises have been developed. However, the state-of-the-art models are not generalizable across populations and probably will not work in the Ugandan context because they have not been implemented with capabilities to eliminate racial discrimination in face recognition. This study therefore developed a deep learning model for masked and unmasked face recognition based on local context. The model was trained and tested on 1000 images taken from students of Kabale University using a Nikon d850 camera. Machine learning techniques such as Principal Component Analysis, Geometric Feature-Based Methods, and double threshold techniques were used in the development phase while results were classified using CNN pre-trained models. From the results obtained, VGG19 achieved a higher accuracy of 91.2% followed by Inception V 3 at 90.3% and VGG16 at 89.69% whereas the developed model achieved 90.32%.