分析血液細胞樣本在醫療診斷與追蹤疾病上的應用事關重要，尤其是血液中白血球細胞的密度，使我們更了解免疫系統的狀態以及可能面臨的潛在風險，當白血球細胞急遽增加時，通常代表著身體受到抗原的影響。而全血細胞計數(Complete Blood Count, CBC)的通常是使用手工計數與血液分析儀，但隨著硬體設備與機器學習演算法的效能提升，使用深度學習來進行血液成分的分類與計數逐漸成為主流方法。在本論文提出改進基於Mask R-CNN的實例分割方法，來檢測白血球細胞。為了提高模型準確度，調整anchor box的比例，使模型更適應血球細胞的大小與形狀，並加入了INMS演算法來修正重疊的細胞區域。我們所使用的資料庫為ALL-IDB，做為深度學習模型的訓練集與測試集，來驗證本論文所提的方法，其中包含108張血液影像和約39000個血球細胞。最終經實驗結果顯示，我們所提出的方法最高在F1分數為69.5%與AP值為91.9%，與原始的Mask R-CNN模型在F1分數為58.5%與AP值為65.1%相比，獲得更高的效能分數。

Analyzing blood cell samples is important for medical diagnosis and tracking of diseases, especially the density of white blood cells in our blood stream, so that we can better understand the state of the immune system and the potential risks that we might be facing. When the number of white blood cells increases rapidly, it usually represents the body is affected by an antigen. The complete blood count (CBC) usually uses manual counting and hematology analyzer. With advancement in the field of hardware equipment and machine learning algorithms improves, deep learning is used to classify and count blood components gradually become the mainstream method. We proposed an instance segmentation method to detect white blood cells, all of which are developed based on Mask R-CNN. In order to improve the accuracy of the model, the scales and aspect ratios of the anchor box are adjusted to fit the size and shape of blood cells. And the INMS algorithm is proposed to correct the overlapping cell area. We use the ALL-IDB database, which is used as the training set and test set of the deep learning model to verify the method proposed in this paper. ALL-IDB database contains 108 blood images and approximately 39,000 blood cells. Experimental results show that our proposed method has a better performance, it’s F1 score = 69.5% and an AP value = 91.9%. Compared with the original Mask R-CNN model, it’s F1 score = 58.5% and the AP value = 65.1%.

[1] N. H. Harun, A. S. Abdul Nasir, M. Y. Mashor and R. Hassan, “Unsupervised segmentation technique for acute leukemia cells using clustering algorithms,” vol. 9, 2015.
[2] D. Umamaheswari, S. Geetha, “Segmentation and classification of acute lymphoblastic leukemia cells tooled with digital image processing and ML techniques,” ICICCS, 2018.
[3] D.-J. Yu et al, “Mixed pooling for convolutional neural networks,” 364-375. 10.1007/978-3-319-11740-9_34, Springer, Cham, 2014.
[4] S. Ioffe et al, “Batch normalization: accelerating deep network training by reducing internal covariate shift,” Int. Conf. Machine Learning, 2015.
[5] Alex Krizhevsky, Ilya Sutskever and Geoffrey E. Hinton, “Imagenet classification with deep convolutional neural networks,” In NIPS, pp. 1106-1114, 2012.
[6] D. Kingma and J. Ba, “Adam: a method for stochastic optimization,” arXiv: 1412.6980, 2014.
[7] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich, “Going deeper with convolutions,” arXiv: 1409-4842, 2014.
[8] Kaiming. He, et al. "Deep residual learning for image recognition." Proceedings of the IEEE conference on computer vision and pattern recognition, 2016.
[9] Ross Girshick, Jeff Donahue, Trevor Darrell, Jitendra Malik, “Rich feature hierarchies for accurate object detection and semantic segmentation,” cs.CV, 2014.
[10] Uijlings, J. R R, Sande V D, et al. “Selective search for object recognition,” International Journal of Computer Vision, 2013.
[11] R. Girshick. “Fast R-CNN,” arXiv:1504.08083, 2015.
[12] Rasmus Rothe, Matthieu Guillaumin, Luc Van Gool. “Non-maximum suppression for object detection by passing messages between windows,” ACCV, 2014.
[13] S. Ren, K. He, R. Girshick, and J. Sun. “Faster R-CNN: towards real-time object detection with region proposal networks,” In NIPS, 2015.
[14] Kaiming He, Georgia Gkioxari, Piotr Dollar, Ross Girshick. “Mask R-CNN,” cs.CV, 2018.
[15] R. Priyatharshini, Aswath Ram. A.S, R. Shyam Sundar, G. Nethaji Nirmal. “Real-time object recognition using region based convolution neural network and recursive neural network,” IJRTE, 2019.
[16] Tsung-Yi Lin, Piotr Dollar, Ross Girshick, Kaiming He, Bharath Hariharan, Serge Belongie. “Feature pyramid networks for object detection,” cs.CV, 2017.
[17] A. P. Sahastrabuddhe, “Counting of RBC and WBC using image processing: a review,” International Journal of Research in Engineering and Technology, 2016.
[18] D. Cruz, et al. “Determination of blood components (WBCs RBCs and Platelets) count in microscopic images using image processing and analysis,” IEEE HNICEM, 2017.
[19] V. D. Dvanesh, P. S. Lakshmi, K. Reddy and A. S. Vasavi, “Blood cell count using digital image processing,” ICCTCT, 2018.
[20] R. Ma, Y. Liang and Y. Ma, “A self-adapting method for RBC count from different blood smears based on PCNN and image quality,” IEEE BIBM, 2016.
[21] V. Aparna, T. V. Sarath and K. I. Ramachandran, “Simulation model for anemia detection using RBC counting algorithms and watershed transform,” ICICICT, 2017.
[22] S. S. Savkare, A. S. Narote and S. P. Narote, “Automatic blood cell segmentation using K-mean clustering from microscopic thin blood images,” VisionNet'16, 2016.
[23] D. Umamaheswari, S. Geetha. “Segmentation and classification ofacute lymphoblastic leukemia cells tooled with digital image processing and ML techniques,” ICICCS, 2018.
[24] Elnaz Mohammadi, Mahdi Orooji. “An unsupervised and supervised combined approach for white blood cells segmentation,” International Iranian Conference on Biomedical Engineering (ICBME), 2018.
[25] Rohit Agrawal, Sachinandan Satapathy, Govind Bagla, K Rajakumar. “Detection of white blood cell cancer using image processing,” ViTECoN, 2019.
[26] Ario Sadafi, Martin Radolko, Iosif Serafeimidis, Steffen Hadlak. “Red blood cells segmentation: a fully convolutional network approach,” IEEE, 2018.
[27] Jonathan Long, Evan Shelhamer, Trevor Darrell.” Fully convolutional networks for semantic segmentation,” cs.CV, 2015.
[28] T. Tran, O. Kwon, K. Kwon, S. Lee and K. Kang. “Blood cell images segmentation using deep learning semantic segmentation,” IEEE ICECE, 2018.
[29] Gan Fang, Yang suhua, Jiang shaofeng. “Detection of white blood cells using YOLOV3 network,” IEEE International Conference on Electronic Measurement & Instruments, 2019.
[30] Bhagyeshri Darane, Prathamesh Rajput, Yogesh Sondagar, Prof. Reeta Koshy. “Recognizing presence of hematological disease using deep learning,” proceedings of the third international conference on I-SMAC, 2019.
[31] Najmeddine Dhieb, Hakim Ghazzai, Hichem Besbes, Yehia Massoud. “An automated blood cells counting and classification framework using Mask R-CNN deep learning model,” ICM, 2019.
[32] The histology guide. https://www.histology.leeds.ac.uk/blood/blood_wbc.php
[33] Acute lymphoblastic leukemia image database for image processing. https://homes.di.unimi.it/scotti/all/
[34] Image polygonal annotation with python，labelme. https://github.com/wkentaro/labelme
[35] Common objects in context (COCO), https://cocodataset.org/#home
[36] Philipp Grabel, et al. “Circular anchors for the detection of hematopoietic cells using retinanet,” ISBI,2020.