Automatic Identification Of Human Biological Cells In Microscopic Fecal Specimens

Feces are the main part of human secretions. The biological cells in feces reflect the pathological changes in the physical condition. The automatic detection method of biological cells in blood or urine has been mature, but fecal samples do not apply this method because the fecal sample has too many impurities. Traditional Fecal biological cell detection is performed by manual operation which is unstable because it extremely depends on the state of individual inspectors. To resolve the issue of a lack of trained personnel and achieve real-time detection, the automation of routine fecal examinations is necessary. To our knowledge, it is the first report that automatic identification of human biological cells on microscopic fecal specimens in the world.With the development of photoelectric technology and computer image processing technology, the digital image processing has been applied to cell recognition step by step. In this thesis, an automated method is proposed to detect biological cells in microscopic fecal specimen images by the way of digital image processing techniques. In order to ensure that the erythrocyte and leukocyte can be detected separately, the sequence of cell detection is specified.Because of the complex environment in human's fecal, segmentation methods such as threshold segmentation and edge segmentation are introduced. By evaluating these segmentation algorithms, mathematical morphological segmentation algorithm is used in this thesis.In this thesis, an automated method is proposed to detect erythrocytes in microscopic fecal specimen images by the way of digital image processing techniques and fuzzy pattern recognition. We use the method combining fuzzy C-means clustering and BP neural network. The fuzzy clustering algorithm can obtain the membership degree of the extracted samples and the standard erythrocyte,The BP neural network can fit the membership function to calculate the degree of membership. In front of the detection, we detect one image and five images in different focal plane which are added for further confirmation in the final step.The shape of leukocyte is more stable than the erythrocyte. In this thesis, the characteristics of leukocyte was extracted, and we detecting the leukocyte by the judgment of the characteristic value. In front of the detection, we also use the five images in different focal plane to detect the leukocyte.In the end, the detection speed of automated method is raised about five times, the performance is tested and analyzed.The automatic detection method has very good results, the true positive rate up to 98% and the false positive rate is also reduced to 9%. This method only takes 16 seconds to detect a single sample, which meets the requirement of the real time detection of the hospital.