About Us   |   Contact Us   |  
Submission  

YOLOv8 for Real-Time Tuberculosis Detection from Low-Resolution Images Using Smartphone Cameras

DOI : https://doi.org/10.36349/easjacc.2025.v07i05.007
PDF
HTML
XML

Tuberculosis (TB) remains a pressing global health issue, requiring timely and accurate diagnosis to prevent its spread and ensure effective treatment. In this study, we explore the po- tential of deep learning and computer vision to enhance TB detection using readily accessible tools like smartphone cameras. Specifically, we leverage the YOLOv8 object detection algo- rithm to analyze images of microscopic slides stained for TB, captured via smartphones. The dataset used in this study consists of 1,224 annotated images sourced from Roboflow, divided into training (861 images), validation (244 images), and test (119 images) sets. Our YOLOv8 model was trained to identify TB bacteria within these images, employing various data augmentation techniques to improve generalization. The model was trained over 100 epochs, and we applied hyperparameter tuning to optimize performance. The training process took approximately 0.826 hours. After training, the model achieved a precision of 72.7%, a recall of 78.7%, and a mean average precision (mAP) of 82.7% at an IoU threshold of 0.5. Additionally, the overall mAP (IoU from 0.5 to 0.95) was 41.5%. The final model size, after stripping the optimizer, was reduced to 22.5MB. These results demonstrate that YOLOv8 is well-suited for TB detection, offering reliable performance with potential real-world applications, especially in remote areas where access to specialized diagnostic equipment is limited. By incorporating YOLOv8 into a smartphone- based diagnostic tool, we propose a more accessible solution for TB detection that could as- sist healthcare workers in resource-constrained settings. This approach not only increases the speed of TB diagnosis but also helps address the challenges associated with traditional meth- ods, which are often time-consuming and require trained personnel. Our findings suggest that YOLOv8, combined with the ubiquity of smartphones, can play a crucial role in advancing TB diagnostics globally.

TOP EDITORS

OPEN ACCESS JOURNALS

Professor Thomas Count Dracula, MD, PhD

Distinguished Professor of Haematology Head — Experimental, Historical & Sensory Haematology Vlad the Impaler University, Wolf’s Lane, Wooden Stakes Grove 666, Transylvania.

BEST AUTHOR

Of The Month

TRACK YOUR ARTICLE

Enter the Manuscript Reference Number (MRN)
Get Details

Contact us


EAS Publisher (East African Scholars Publisher)
Nairobi, Kenya


Phone : +91-9365665504
Whatsapp : +91-8724002629
Email : easpublisher@gmail.com

About Us


EAS Publisher (East African Scholars Publisher) is an international scholar’s publisher for open access scientific journals in both print and online publishing from Kenya. Its aim is to provide scholars ... Read More Here

*This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.

© 2020, All Rights Reserved | SASPR Edu International Pvt. Ltd.

Developed by JM