Diagnosis of Mycobacterium Tuberculosis via K-Means and Canny Edge Detection

Tuberculosis is one of the leading causes of death with 2,000 death cases in Malaysia each year. This raises the importance of quick and accurate diagnosis of tuberculosis to ensure timely treatment. Most low-middle-income countries, including Malaysia, have been adopting sputum smear microscopy to detect tuberculosis, but such method is time consuming, labour intensive and inaccurate. Other detection methods such as mycobacterium culture and sensitivity test, and chest x-ray offer higher accuracy, but are much more expensive. Therefore, our study employed image processing technique on microscopic image of patients’ sputum samples to substitute human labour for rapid and accurate tuberculosis diagnosis. K-means algorithm was used as an image segmentation technique, followed by image classification using Canny Edge Detection algorithm to obtain the bacilli count. The bacilli count per sputum sample was used as a parameter to conduct tuberculosis diagnosis by World Health Organisation standard. The proposed method combining K-means and Canny Edge Detection algorithm achieved 88.9% diagnosis accuracy. On-going work has been carried out to assess the feasibility of utilising Machine Learning Model Builder from ML.net to decide the best algorithm for acid-fast bacilli counting based on the patients’ sputum samples training data.