CC BY
0AN ANALYSIS OF THE USE OF THE YOLO ALGORITHM IN THE DIAGNOSIS OF
BLOOD CELL IMAGES 1Eshmuradov D.E,2Iskandarova Sayora,3Tulaganova F.K,
1,2TATU Associate professor, 3TATU doctoral student https://doi.org/10.5281/zenodo.10723507
Abstract. Peripheral smear is a method that allows the analysis of blood with the help of a microscope after staining the blood liquid on a glass slide. The morphological appearance and multiplicity of some blood elements are examined by this method. This examination is preferred for evaluating the response to treatment as well as providing information about the diagnosis and severity of diseases. Since manual examination of blood elements by operators contains many humanistic and environmental parameters, it is time-consuming and error-prone. Instead of this approach which does not offer a standard accuracy rate, it is planned to create a quick, cost-effective and fully automatic computer-aided system. In this study, the examination of ALL (Acute Lymphoblastic Leukaemia) malignancy which is one of the main types of leukaemia by peripheral smear method will be handled
Keywords: X-ray films, Machine Learning, diagnosis, performance, algorithms, differentia, dental panoramic.
Introduction Acute lymphoblastic leukaemia is a hematic disorder known as cancer of the white blood cells. It is occurred by the excessive production of malignant and immature white blood cells known as lymphoblast or blast produced in the bone marrow. This malignancy, which generally affects young children and adults over 50, spreads rapidly to the bloodstream and vital organs. Therefore, early diagnosis is the basic principle for the sustainability of vitality.
In the first stage of ALL cancer, general symptoms such as weakness, weight loss, loss of appetite, nosebleeds, bone or joint pains and shortness of breath are observed. However, in the later stages, blast cells that are overproduced in the bone marrow begin to be seen in the peripheral blood. Evaluation of peripheral blood film is part of the diagnostic process.
This study is planned to perform the analysis of blood cells in a fast and cost-effective manner. And also, it is aimed to design a fully automatic system that assists the operators in obtaining the standard accuracy rate. The flow diagram of the current study focusing on the separation and counting of immature white blood cells in the blood from other elements is presented in Figure 1.
Methodology. In the present study, it is planned to distinguish ALL patients from healthy individuals. For this purpose, analyzes are made on the microscopic images of peripheral blood samples obtained from the website https://homes.di.unimi.it/scotti/all/. All images in the public and free ALL-IDB dataset are captured at different magnifications and densities using the Canon PowerShot G5 camera. Assessments are provided using images from the ALL-DIB 1+ALL-IDB2 datasets.
YOLO (You Only Look Once) Approach
YOLO is a deep learning network that detects and classifies blast cells in blood samples captured by the microscope. The inputs given to the algorithm are the data obtained with different magnification rates from areas with varied densities of blood elements. Version 4 of Yolo is preferred for accurate distinguishing and counting of blast cells segmented on these data.
Figure 1. Flow chart of the presented study
The YOLOV4 algorithm developed by is an enhanced version of YOLOV3. The backbone part is the CSPDarknet53 structure, in which CSPNet and Darknet53 architectures are used together. This structure which plays a role in feature extraction reduces computation and memory costs and improves the learning ability of CNN. The neck part of the YOLOV4 algorithm consists of SPP and PAN approaches and the head part consists of the YOLOV3 algorithm.
SPP is a module that improves the receptive field and consists of Spatial Pyramid Matching (SPM). SPM (Spatial Pyramid Matching) divides the input image as SXS into a grid to obtain the spatial pyramid. It is then integrated into CNN by SPP. The PAN structure is used to collect parameters from different backbone levels.
YOLOV4 being a one-step method produces output by calculating the coordinates and category probability of the given input. The target detection process is presented in Figure 2.
Figure 2. Target detection process for YOLOv4 algorithm.
The YOLO algorithm, in which the image is considered as a whole, separates the data into SXS coordinates. It estimates the bounding box coordinates for the relevant object in the image and evaluates the probability that the object in these boxes is a blast cell. The probability of whether a blast cell is present or not is available for all grids, and aspect ratios are estimated based on the entire image.
The rates of loss value obtained as a result of the training performed on the ALL-IDB 1+ALL-IDB2 datasets with the YOLOV4 algorithm are presented in Figure3.
ILO
«a.ft
• o
M-O
xw an MS _tan IMS MM) «IM MM MM M
rwrmt •«< t*«« • OJIî* H»» »< I in * WO «pyrai lira* Ml « O > ■ l»ur«
Figure 3. Detection and counting process on ALL-IDB 1+ALL-IDB2 dataset
Table1: A comparison on studies using the ALL-IDB dataset
A comparison on studies using the ALL-IDB dataset
Author Aim of study Method Dataset Results
Jha, K. K, & Dutta, H. S. 2019 Leukaemia detection modul from the blood smear images MI+LDP+SCA based deep CN classifier ALL-IDB2 Accuracy is % 98,7
Al-jaboriy, S.S. Sjarif, N.N. A. Chuprat, S. & Abduallah, W. M. To develop a new and improved technique for leukocyte cell segmentation GA+ANN ALL-IDB1 Identified blast cell accuracy is %97
Al-Tahhan, F.E. Fares, M.E. Sakr, A.A. & Aladle, D.A. 2020. To identify automatically the ALL subtypest (L1/L2/L3) KNN/SVM/ANN ALL-IDB2 Maximum accuracy is obtained as % 100 by quadratic SVM
Safuan, S.N. M. Tomari, M.R.M. Zakaria, W.N.W. Mohd, M.N.H. & Suriani, N.S. 2020. To differentiate lymphoblast and non-lymphoblast cells and examine ALL disease VGG/GoogleNet/Alexnet ALL-IDB2 (60 images) Maximum accuracy is obtained as % 99,13 by VGG
Sahlol, A.T. Kollmannsberger P. & Ewees, A.A. 2020 White blood cell leukaemia image classification SESSA+VGGNet ALL-IDB2 Accuracy is % 96,11
Anilkumar, K.K. Manoj, V.J. & Sagi, T.M. 2021. Automated identification of leukaemia Transfer Learning Architectures ALL-IDB1 ALL-IDB2 Maximum accuracy is % 100 both ALL-D-IDB1 and ALL-IDB2 dataset
This study aims to evaluate ALL disease using the peripheral blood smear technique. As a result of the study, the output will be produced in 3-4 seconds with real-time automatic diagnosis provided on digitized peripheral blood smear images. With this study on the identification and counting of blast cells from hundreds of cells, the severity of the disease will be evaluated, the right treatment protocol will be selected and the response to the treatment will be monitored. It is expected that this study will provide helpful ideas to physicians.
REFERENCES
1. Pal, R., Poray, J., & Sen, M. (2017, May). Application of machine learning algorithms on diabetic retinopathy. In 2017 2nd IEEE International Conference on Recent Trends in Electronics, Information & Communication Technology (RTEICT) (pp. 2046-2051). IEEE.
2. Hou, X., Li, Q., Wang, Q., Zhou, M., & Liu, H. (2016, October). An improved SAM algorithm for red blood cells and white blood cells segmentation. In 2016 9th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI) (pp. 651-655). IEEE.
3. Rahman, I. U., Sulaiman, M., Alarfaj, F. K., Kumam, P., & Laouini, G. (2021). Investigation of non-linear MHD Jeffery-Hamel blood flow model using a hybrid metaheuristic approach. IEEE Access, 9, 163214-163232.
4. Guedri, N., & Gharbi, R. (2022, December). Detection of the breast cancer cells by a smart camera connected to a cellular Microscopy. In 2022 IEEE 21st international Ccnference on Sciences and Techniques of Automatic Control and Computer Engineering (STA) (pp. 79-83). IEEE.
5. Nugroho, H. A., Saputra, W. A., Permanasari, A. E., & Murhandarwati, E. E. H. (2017, August). Automated determination of Plasmodium region of interest on thin blood smear images. In 2017 international seminar on intelligent technology and its applications (ISITIA) (pp. 352-355). IEEE.
6. Rey-Barroso, L., Roldán, M., Burgos-Fernández, F. J., Gassiot, S., Llobet, A. R., Isola, I., & Vilaseca, M. (2021, June). Evaluating Confocal Microscopy as a Tool to Diagnose Red Blood Cell Diseases. In The European Conference on Lasers and Electro-Optics (p. ch_p_20). Optica Publishing Group.
7. Nabeel, P. M., Manoj, R., Abhidev, V. V., Joseph, J., & Sivaprakasam, M. (2020, July). High-Throughput Vascular Screening by ARTSENS Pen During a Medical Camp for Early-Stage Detection of Chronic Kidney Disease. In 2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) (pp. 2752-2755). IEEE.
8. Eshmurodov D. Iskanderova S. Tulaganova F. (January 29-31, 2024) Algorithms for detection of cells in blood images
9. Eshmuradov D., Ismailov O., Magrupova M. METHODS AND MEANS OF DIGITAL PROCESSING OF BIOELECTRIC SIGNALS //Science and innovation. - 2023. - T. 2. - №. A2. - C. 84-88.
10. O.Ismailov, M.Fozilova, S.Mirzakhalilov, M.Ismoilov. "Primary diagnosis of cardiovascular diseases using artificial intelligence methods." International scientific and practical conference "BIG DATA, artificial intelligence and e-commerce in the development of the digital economy: problems, challenges, solutions." Tashkent, 2023, 162 - 166 p.
