Adegoke, Faith Omolara and Covenant University, Theses (2023) DEVELOPMENT OF AN AUTOMATED MALARIA DETECTION SYSTEM USING DEEP LEARNING MODELS. Masters thesis, Covenant University Ota.
![[img]](http://eprints.covenantuniversity.edu.ng/style/images/fileicons/application_pdf.png) |
PDF
Download (220kB) |
Abstract
Malaria remains a public health concern, prompting intensive research into computer-aided diagnosis using machine learning models. However, the effectiveness of these models is hindered by the presence of variabilities in clinical practices, especially in medical imaging. These issues, such as demographic differences among patients, diverse staining methods, and variations in devices, coupled with the absence of standardized medical protocols, present considerable obstacles in achieving optimal model performance. This study aims to develop a malaria detection system using deep learning models. Existing malaria image datasets were curated by doing a systematic literature review of papers published in this domain from 2015-2023. Thirteen (13) datasets were retrieved following this process. The key artifacts classified are the infection status, parasite species, type of stain, type of smear, and optical train. Python programming was used in implementing the deep learning models for the classification of the identified artifacts. Getting this information about an image dataset will ensure standardized approaches to diagnosis and research, leading to more reliable and comparable data across different settings and studies. The performance evaluation metrics used include recall (sensitivity), accuracy, precision, and F1-score. A ten-fold cross validation was also done. The models were evaluated on single set and combined sets (to increase statistical power) to compare their performance. The best performing model for infection status is VGG19 on the single set, RESNET50 performed best on the single set for Species classification, for classifying smears, VGG19 performed best on the combined set, RESNET50 performed best on the single set for classifying stains and for classifying optical train, VGG19 had the best performance on the combined set. A prototype web application for the prediction of these artifacts was developed using the Python Flask micro-framework. The best performing models were loaded to the web application. When deployed, it will provide a user-friendly platform for medical professionals and researchers alike.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Malaria, Image classification, Deep Learning, Computer-aided Diagnosis, Blood Smear Images |
| Subjects: | Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QH Natural history > QH301 Biology |
| Divisions: | Faculty of Engineering, Science and Mathematics > School of Electronics and Computer Science |
| Depositing User: | nwokealisi |
| Date Deposited: | 11 Sep 2023 10:43 |
| Last Modified: | 11 Sep 2023 10:43 |
| URI: | http://eprints.covenantuniversity.edu.ng/id/eprint/17293 |
Actions (login required)
 |
View Item |
