PAUL, NAOMI ADAEZE (2026) DEVELOPMENT OF AN AUTOMATED KIDNEY STONE DETECTION SYSTEM USING DEEP LEARNING ARCHITECTURE. Other thesis, Godfrey Okoye University, Enugu.
|
Text
AI Project.pdf Download (3MB) | Preview |
Abstract
Kidney stone disease or Nephrolithiasis is a major global health burden that is associated with significant morbidity and complication if left undiagnosed, untreated or misdiagnosed, such as urinary obstruction, infection, and chronic kidney disease. Current diagnostic workflows are based on subjective interpretation of computed tomography (CT) scan images by radiologists, but the process is time-consuming, and the results may vary between different observers and be wrong. In this paper, a web-based deep learning system for the automated detection of kidney stones based on the YOLOv8 object detection architecture is developed and tested. The system consists of a YOLOv8 model to extract features and locate the objects, a FastAPI backend for processing inferences, and a web application using HTML and CSS for a responsive interface. The model was trained on a curated Kaggle dataset using data augmentation, pixel normalization, and resizing to 640×640 pixels to improve the model's capabilities to generalize. Experimental results showed good performance, with a precision of 0.91, recall of 0.89, F1-score of 0.90, and a mean Average Precision (mAP) of 0.93. The inference time of 0.35 s per image was a testament to real-time performance, which is crucial for clinical workflows. The integrated system allows the user to upload medical images and get the real-time detection results (bounding box coordinates and confidence scores).The web interface features login authentication, landing page, dashboard, patient information input, drag-and-drop uploading of images, feedback on the loading screen, clinical recommendation on results, positive-negative case inference views and history tracking. It is not meant to replace clinical skills, but rather it can be considered a decision support system which helps improve the diagnostic accuracy and shorten the turnaround time. Adaptations to other disease states and hospital information systems could be extensions.
| Item Type: | Thesis (Other) |
|---|---|
| Subjects: | Q Science > Q Science (General) |
| Divisions: | Faculty of Engineering, Science and Mathematics > School of Electronics and Computer Science |
| Depositing User: | MICHAEL MADUBUKO |
| Date Deposited: | 22 Jul 2026 10:00 |
| Last Modified: | 22 Jul 2026 10:00 |
| URI: | http://eprints.gouni.edu.ng/id/eprint/5971 |
Actions (login required)
 |
View Item |
