Jimma University Open access Institutional Repository

Automated Detection System for Tuberculosis from Sputum Smear Microscope Images Using Deep Neural Networks

Show simple item record

dc.contributor.author Gedamu, Fikremariam
dc.contributor.author Kwa, Timothy
dc.contributor.author Worku, Abel
dc.date.accessioned 2023-02-10T07:17:51Z
dc.date.available 2023-02-10T07:17:51Z
dc.date.issued 2022-08-12
dc.identifier.uri https://repository.ju.edu.et//handle/123456789/7633
dc.description.abstract Tuberculosis (TB) is an infectious disease caused by Mycobacterium tuberculosis bacteria and most often affects the lung. It was the leading cause of death worldwide from a single infectious agent until the COVID-19 pandemic, ranking above HIV/AIDS. Sputum smear microscopy is the main TB diagnosis and treatment monitoring tool and the procedure is observing the sputum smear through the microscope for tuberculosis bacteria. It is a time-consuming procedure and highly prone to human error. Traditional machine learning techniques were implemented to classify the sputum microscope image whether it is positive or negative for tuberculosis bacteria. It is well known that image processing and traditional machine learning techniques have problems with de tecting and counting bacilli and with bacilli classification. This thesis is mainly comprised of the classification and detection of tuberculosis bacteria from a sputum microscope images. Classifica tion of the sputum microscope image using a Convolutional Neural Network was done before the object detection task to reduce the computational cost. The CNN was trained on 7905 sputum microscope images with an image size of 256x256 and achieved accuracy, sensitivity, and speci ficity of 98%,98%, and 97% respectively. After classification, the positive sputum microscope image was passed to object detection using TensorFlow object detection API which is an open source framework based on Google TensorFlow which allows us to create, train and deploy object detection models. SSD-MobileNet-v1-fpn, Faster RCNN-Inception-v2, and RFCN-ResNet-101 were trained on datasets of 1440 images with a size of 640x640 to select the best among them. The results suggested that the performance of Faster RCNN Inception v2 is the best with mean average precision score of 93%, followed by RFCN-ResNet-101 with 92% and SSD-MobileNet-v1-fpn with 90% of mean average precision. The results provided substantial evidence that the Faster RCNN model is the most accurate and suitable model for Tuberculosis Bacilli detection on sputum smear microscope images en_US
dc.language.iso en_US en_US
dc.subject Classification en_US
dc.subject Detection en_US
dc.subject FRCN en_US
dc.subject RFCN en_US
dc.subject SSD en_US
dc.subject Tuberculosis en_US
dc.subject regimen en_US
dc.subject Sputum smear microscopy en_US
dc.title Automated Detection System for Tuberculosis from Sputum Smear Microscope Images Using Deep Neural Networks en_US
dc.type Thesis en_US


Files in this item

This item appears in the following Collection(s)

Show simple item record

Search IR


Browse

My Account