Research Of Ultrasound In The Diagnosis Of Primary Hepatic Carcinoma Applying Of Intelligent Information Processing Technology

Objective: The present study combines the principal component analysis(PCA) with BP neural network(BPNNs),which are belong to intelligent information processing technology, in order to establish a diagnosis mathematical model called PCA-BPNNs model for assisting sonographers with diagnosis of hepatocellular carcinoma(HCC) in ultrasonography. In the study, we evaluate the theoretical value and diagnostic performance of the model expecting it to improve the accuracy of diagnosis of HCC in ultrasonography.Materials and Methods: 1. Data Collection:The study collects 350 focal liver lesions of ultrasound images and clinical data, comprising of 184 HCCs, 83 hemangiomas, 42 metastases, 19 micronodular cirrhosis,13 focal nodular hyperplasias, 9 hepatic abscess,in which 125 HCC cases as the training set, and 225 cases Including the remaining 59 HCC cases and 166 cases of other five kinds of focal liver lesions as the testing set. 2. Establishment of PCA-BPNNs based model for diagnosis of HCC in Ultrasonography: In the first step, The PCA based image fusioning technique is used for the image preprocessing. After the extraction and quantification of sonographic and clinical data, the study use PCA to extract the characteristic parameters of HCC. Secondly, we use the characteristic parameters to establish diagnostic model, then train and emulate it. 3. The study compare the simulation results with the pathology test results. 4. Diagnostic performance evaluation of the model: Make ROC curves for PCA-BPNNs based model’s diagnosis, PCA based image fusioning technology’s diagnosis, BPNNs based model’s diagnosis and sonographers’ diagnosis respectively and compare their diagnostic performance using the size of the area under curve.Results: 1. PCA based image fusioning results indicate that the images are clearer and its characteristics are more prominent after preprocessing. 2. A total of 13 feature parameters of HCC extracted by PCA including: irregular lesion shapes, lesion boundary is not clear, halo ring signs, mosaic signs, satellite nodules, blood signals inside or around the lesions, the lesion supplies by arterial and venous blood, RI≥0.6, portal embolus, hilar lymphadenopathy, HBV, cirrhosis of liver, AFP≥400μg/L. 3. A total of 16 feature parameters of HCC extracted by sonographer including:liver parenchymal echo uneven, rough liver parenchymal echo, pseudopodia signs and 13 parameters extracted by PCA. 4. Among the results of PCA based image fusioning technology method, BPNNs based model and sonographers, PCA-BPNNs based model’s results is the best, which has a diagnosis sensitivity of 93.22%, specificity of 94.58% and accuracy of 94.22%. 5. These four methods have their area under the ROC curve respectively: PCA-BPNNsbased model of 0.939, BPNNs based model of 0.919, PCA based image fusioning technology of 0.819 and sonographers of 0.814, which shows that the diagnostic performance of PCA-BPNNs based model is the best.Conclusions: 1. In our research, a PCA-BPNNs HCC ultrasonic diagnosis mathematical model can be established by using PCA image fusion and BPNNs. 2. The PCA-BPNNs based model is help to improve the accuracy of the diagnosis of HCC in ultrasonography. 3. The diagnostic performance of PCA-BPNNs based model is better than that of a single application of BPNNs or PCA image fusion in ultrasonic diagnosis of HCC.