| Wireless Capsule Endoscopy (WCE) system can directly image the humangastrointestinal (GI) tract, and allows clinicians to directly view the lesions anddiagnose the GI tract diseases noninvasively. The usage of WCE is convenientand painless, and the entire GI tract is examined without a dead zone. TheWCE is applied more and more in clinical examinations based on these assets,and the research on technologies concerning WCE has become popular in themedical device industry at home and abroad. As a novel technology, theexisting WCE systems can not yet satisfy the clinical demands, because of thelow quality image, low frame rate and limited working time. In the meaningtime, WCE will generate a large number of images in one examination of apatient. It is, therefore, very laborious and time-consuming to review the WCEimages and hard to find the lesions of diseases, thus limiting the application ofWCE. Aimed at these problems, with the supports from the National HighTechnology Research and Development Program of China（863Program）, theNational Natural Science Foundation of China and the Science and TechnologyCommission of Shanghai Municipality, this thesis has carried out deeplyresearch on the technologies of WCE system and the intelligent bleedingdetection from WCE images. Two kindsof WCE systems have been developedin this thesis. The concept of color vector similarity coefficient has beenproposed to quantitatively measure the similarity degree of different colors, andthe calculation formula has been deduced. And then a new bleeding detectionfrom WCE images based on the color vector similarity coefficient has beenimplemented. Using differential evolution (DE) to improve the basicprobabilistic neural networks (PNN), a new PNN with different smoothingparameters in each neuron has been built, and then the intelligent bleedingdetection based on the improved PNN has been implemented.In the existing WCE systems the GI tract images are transmitted withanalog signal, and then the analog signal is converted back into digital imagesin the receiving box. The analog signal is susceptible to interference, inaddition that some information has to be lost during the two times of A/D andD/A conversion. As a consequence, the WCE images can not yet satisfy thedemands of clinical diagnose. A novel JPEG-based digital WCE system hasbeen developed in this thesis, which is constituted of JPEG-based capsule endoscope, image receiving box, image working station and the applicationsoftware. The JPEG-based capsule endoscope is12mm in length and28mm indiameter, which is small enough to be swallowed for patients. After beingswallowed, the capsule endoscope travels through the entire GI tract withnatural peristalses. During the course the capsule endoscope has the ability tocapture GI tract images with the resolution of320240, and the ability tocompress the Bayer images into JPEG image format, and the compression ratecan be adjusted through the quantization table. The compressed JPEG imagesare directly transmitted out wirelessly and digitally, and so two times of D/Aand A/D conversion are saved. The digital JPEG images are received and storedby the outer receiving box which is tied to the patient’s waist. Afterexamination, the images are downloaded into the image workstation for playand diagnosis. Compared with the existing WCE systems with analog signal,digital signal is stronger against the interference, and so the images are clear,which will improve the diagnosis rate.The existing capsule endoscope is powered by a cell button, which can keepworking continuously for5hours, and the frame rate is about2f/s. The limitedworking time and low image frame rate limit the wider application. Because2f/s is obviously not enough for diagnose the details of GI tract. Furthermore,for a patient with GI tract disease, WCE travels slowly and it needs greatlymore than8hours to diagnose the entire GI tract, as a consequence, the cell button often runs out of power while the small intestine is still not examined. Inthe meaning time, the cell button in human body is a kind of underlying threat.A novel NTSC video WCE system based on wireless power supply isdeveloped in this thesis which is constituted of video capsule endoscope,wireless electric power transmitting device, image receiving box and imagework station. The capsule endoscope is10mm in diameter and30mm in length,and has the ability to capture GI tract images in human body, and the ability tocode the images into NTSC video with the frame rate of30f/s. The video signalis transmitted out of patient’s body wirelessly and received by the imagereceiving box. In the receiving box, the NTSC video is encoded into MPEGⅡformat digital video file and saved in it. After the examination being finished,the digital video file is downloaded into the image work station to be played bythe special application software and diagnosed by the clinician. The receivingbox also has the ability to deliver the NTSC video signal directly onto theworkstation, and NTSC video is played real-time. So the clinician cansupervise the examination and diagnose real-time. The wireless power supplysystem is constituted of the outside wireless power transmitting device and theinner wireless power receiver subsystem embedded in the capsule endoscope.The outside wireless power transmitting device is constituted of signalgenerator, driver circuit model and a pair of Helmholtz transmitting coils (TC).Helmholtz TC can generate uniformed changing magnetic field in the special space. The wireless power receiving subsystem is constituted of a3dimensional receiving coil (RC) and a rectifier voltage regulator. Thecollaboration between Helmhoz TC and3D RC guarantees the video capsuleendoscope can receive electric power at any posture anywhere, and theproblems of position and posture stabilities are resolved. Because the videocapsule endoscope is powered wirelessly, the video WCE system can work aslong as needed, which is a real long working time and high speed WCE system.This research work lays a theoretical and technological foundation for theapplication of high speed video WCE system.The bleeding regions in WCE images are analyzed and the features areexacted in this thesis. Bleeding pixels and non-bleeding pixels are grouped intodifferent pattern classes, and there is a certain color region separating thebleeding pattern class from non-bleeding pattern class. Then the color featurescan be used to recognize the bleeding pixels and then recognize the bleedingWCE images. Combining the color representation with the vector calculation,this thesis proposes the color vector similarity coefficients to quantitativelymeasure the similarity degree between different colors, and deduces thecalculation formulas of the similarity coefficients. The color vector similaritycoefficients include chroma similarity coefficient and gray intensity similaritycoefficient. More similar the colors are, bigger the coefficients are, and whentwo color are exact same, both chroma and gray intensity similarity coefficients get their maxima value1. The chroma and gray intensity similarity coefficientscan be used to quantitatively measure the similarity degree between differentcolors, and are great tools for color image processing. The bleeding patternclassifier based on the color vector similarity coefficient is built in this thesis,and combining the seed region growing, the intelligent bleeding detectionalgorithm based on the color vector similarity coefficient is implemented. Theexperiments are designed to test the bleeding detection algorithm, which ismeasured to have the sensitivity of97%and the specificity of90%. Theexperiments show that the algorithm is featured as fast calculation and highsensitivity.Artificial neural network (ANN) is featured as good abilities of adaptiveand self-learning, and is widely applied in variety of pattern recognitionproblems. The bleeding detection expert system (ES) based on ANN is an idealsolution for the bleeding detection in WCE images, and can bring good basisfor the disease lesions detection. The features of bleeding region in WCEimages are extracted in RGB and HSI color spaces to form the feature vectors.Using the feature vectors as the input, a BP neural networks classifier is built,based on which, the intelligent bleeding detection algorithm is implementedthrough programming. The experiment show that the sensitivity and specificityof this bleeding detection algorithm is93%and96%respectively. Probabilisticneural network (PNN) is a feedforward neural network based on the radial basis function and the Bayesian theory. It combines some of the best attributes ofstatistical pattern recognition and feedforward artificial neural networks, andcan produce outputs with Bayesian posterior probabilities. It is also characteredas fast learning and stability, and so is very competent for the nonlinear patternrecognition. Nevertheless basic PNN has same smoothing parameters σ inevery neuron, and so the recognition rate is low. Using the differentialevolution (DE) algorithm, this thesis has improved the basis PNN to holddifferent smoothing parameters σ in each neuron, so the recognition rate isimproved. Using the feature vector of color pixel in HSI and RGB color spaceas the input, the bleeding pattern classifier is built based on the improved PNN,and the intelligent bleeding detection algorithm based on improved PNN isimplemented. The experiment shows that the sensitivity of this algorithm is93.1%and the specificity is85.8%. Compared with BP bleeding detectionalgorithm, the improved PNN bleeding detection algorithm is charactered asfast recognition and stable architecture. The automatic and intelligent bleedingdetection is implemented and will be used in the clinical filed to process theWCE images primarily, and lays the foundation for the intelligent detections ofother kinds of lesions. |
PDF Full Text Request