Image segmentation is one of the most critical steps in computer-aided diagnosis that potentially accelerate leukemia diagnosis. Leukemia is categorized as blood cancer known as a deadly disease. Generally, acute lymphoblastic leukemia (ALL) detection can be done manually by counting the leukocytes contained in the stained peripheral blood smear image using the immunohistochemical (IHC) method. Unfortunately, the manual diagnosis process takes 3−24 hours to complete and is most likely inaccurate due to operator fatigue. An image segmentation method proposed by Vogado can achieve an accuracy of 98.5%. However, this method uses a K-means clustering algorithm that is not optimal for input images containing mostly noise. In this research, fuzzy c-means were applied to solve this problem. The dataset used in this study was ALL-IDB2, which consisted of 260 images, with each image having the size of 257×257 pixels in tagged image file (TIF) format. The initial stage of this method was to divide the ALL-IDB 2 acute leukemia dataset image into cyan, magenta, yellow, key (CMYK) and L*a*b color schemes which then subtract the M component subtracted by component *b. The subtraction results were then splits using the FCM algorithm, resulting in the nucleus and background sections. The output of this method was then evaluated and measured using the metrics accuracy, specificity, sensitivity, kappa index, dice coefficient, and time complexity. The results showed that changing the clustering algorithm in the image segmentation method did not provide a significant change in results; an increase occurred in the specificity and precision metrics with an average of 0.1−0.4%, the execution time also increased by an average of 23.10%. The decrease in the accuracy metric was down to 95.4238%, and the dice coefficient value was 79.3682%. From the explanation above, it can be concluded that the application of the FCM algorithm to the segmentation method does not provide optimal results.
Copyrights © 2022