03 December 2021 Volume :9 Issue :55

Tackling TB Using Automatic Detection

Tackling TB Using Automatic Detection
Mr Mustapha Olayemi Oloko-Oba, PhD candidate in the School of Mathematics, Statistics and Computer Science.

Mr Mustapha Olayemi Oloko-Oba is looking forward to taking part in the Postgraduate Research and Innovation Symposium (PRIS 2021) hosted by UKZN’s College of Agriculture, Engineering and Science (CAES) where he will be presenting on the topic: Ensemble of Convolution Neural Networks for Automatic Tuberculosis Classification.

Oloko-Oba is a PhD candidate under the supervision of Professor Serestina Viriri in the School of Mathematics, Statistics and Computer Science. He holds a BSc (Hons) from Kogi State University, Nigeria, and a Master’s degree in Computer Science from the University of Ilorin, Nigeria. His current research area is image processing (medical), with a specialisation in computer vision.

His research involves Tuberculosis (TB) detection using a computer-aided detection system. Tuberculosis is ranked among the highest causes of death and is most prevalent in developing countries, especially in Africa. According to the World Health Organization (WHO), it is the cause of much economic distress, poverty and vulnerability, with South-East Asia and Africa accounting for about 69% of total cases. This motivated Oloko-Oba to use Deep Learning models to develop an automatic detection system. ‘This system will assist with early diagnosis, correct misdiagnosis, eliminate the bottleneck of skilled radiologists and ultimately avert millions of deaths,’ he said.

With early detection, TB is curable, and millions of deaths could be averted. ‘I want to contribute to addressing a life-threatening disease through meaningful research,’ added Oloko-Oba.

One of the most reliable ways to screen for TB is a chest X-ray; however, its success depends on the interpretation of skilled and experienced radiologists and regions with high TB burdens lack such skills. A computer-aided system can automatically detect TB from chest X-rays. ‘We employed state-of-the-art Deep Learning models through Ensemble learning to automatically detect and classify infected CXR from healthy ones. Our model was trained on the Shenzhen dataset and validated on the Montgomery dataset to improve accuracy and generalisation on new (unseen) datasets as compared to existing models with low sensitivity and accuracy,’ explained Oloko-Oba.

He has participated in several other conferences and highly recommends PRIS that offers ‘constructive criticism from experts that will help shape your research and could attract collaboration.’

Oloko-Oba said he owes his success to his supervisor, Viriri. His also thanked his fellow researchers.

To find out more about Oloko-Oba’s research as well as other CAES researchers at PRIS 2021, visit pris.ukzn.ac.za

Words: Samantha Ngcongo

Photograph: Supplied

author : .
author email : .