Detection and Counting Bacilli Tuberculosis Images using Classical Image Processing and Deep Learning

Tuberculosis (TB) has been known as the top two murdering disease worldwide after HIV/AIDS. Indonesia is the second country with the highest number of tuberculosis. One of the reasons for the high number of cases is that the microscopic sputum smear image analysis process takes a long time. This study compares the detection and counting of bacilli images using classical image processing and deep learning to ascertain the performance of both methods. Steps in classical image processing consist of RGB to HSV converting, HSV thresholding, morphological operation and bacilli counting. YOLOv5 is used as deep learning method in this research to detect and count bacilli. While classical image processing accuracy is 59.59%, YOLOv5 can achieve 86.09% which is better than the previous research. By comparing classical image processing methods and deep learning, it is evident that deep learning techniques can detect and count overlapping bacilli. This result can be further improved by increasing the learning dataset especially the severe TB patient.