Design And Implementation Of Intelligent Diagnosis System For Breast Cancer Ultrasound Image

Objective: Breast cancer,with the world’s highest incidence and the fifth highest mortality rate of cancer,has been highly concerned by the whole society.Medical imaging examination is the most important means of clinical early screening and diagnosis of breast cancer.Accurate imaging diagnosis requires the rich experience of doctors,and the diagnosis process will consume doctors a lot of time and energy.With the development and rapid progress of artificial intelligence technology,intelligent algorithms such as deep learning can extract a large amount of semantic information contained in medical images and have been gradually applied to medical image processing in recent years,showing good technical advantages and application value.Ultrasound imaging is one of the primary choices for breast cancer screening.The metastatic status of axillary lymph nodes plays an important role in the diagnosis and prognosis of breast cancer,and the accurate segmentation of ultrasound images is beneficial to the clinical diagnosis of breast cancer.On the one hand,at present,it is time-consuming and laborious to directly predict axillary lymph node metastasis through ultrasonic image signs,and the accuracy of computer-aided diagnosis based on traditional methods still needs to be improved.The application of artificial intelligence algorithms such as deep learning in its ultrasonic image diagnosis is relatively rare.On the other hand,the application of a medical image intelligent diagnosis system in the hospital also has great potential demand and application value.In conclusion,this paper,based on the deep learning technology,implements breast ultrasound image research intelligent diagnosis system,to realize an intelligent diagnosis of breast cancer-related images of high accuracy,marked for clinical doctors’ interest area,provide diagnostic information,reading to reduce the workload,improve the efficiency of diagnosis,has important research significance and application value.Methods & Results: This paper designs and implements an intelligent diagnosis system of breast cancer based on ultrasonic images.The system not only meets the basic daily work but also has the function of intelligent assisting doctors in clinical diagnosis.In the image intelligent diagnosis function of the system,the segmentation task of the region of interest and the classification task of predicting axillary lymph node metastasis in breast cancer ultrasound images are mainly realized.In the aspect of the segmentation task,the U-Net-MDSC algorithm is designed based on the U-Net algorithm.The algorithm adds the dense hop connection structure to the original network structure to improve the feature extraction ability of the network model and avoid losing more semantic information in the process of downsampling.For experimental data in this paper,ultrasonic image set itself the characteristics of small size,low quality,this article has been done for the sampling structure part pruning operation on the last layers,realized it does not affect the premise of accuracy,improve the training speed of the network model,the parameters and reduce the training time,eventually,the Dice coefficient of the model is 0.903,The results are the highest compared to other similar networks.Given the classification task,firstly,a convolution block is introduced into the breast cancer ultrasound image classification network model for the first time in this algorithm.The structured shape of the convolution block is an inverted bottleneck structure.Secondly,given this kind of ultrasonic image data and two categories of imbalances,fewer studies using the migration will be trained in other large amounts of data of model parameters are transferred to the paper on the experimental data,after parameter adjustment,in this paper,the classification of network model accuracy,sensitivity,and specificity of the results reached 93.8%,94.9%,and 94.8%,respectively,Compared with similar classification algorithms using transfer learning,the result is the best.Finally,the above two tasks are implemented in the Tensorflow deep learning framework,and the best results are achieved in the comparative experiment,which verifies the feasibility and advantages of the two algorithms in this paper.In addition,in the intelligent diagnosis system design,this paper designed for doctors,patients,and administrators three kinds of user demand analysis interfaces,and based on the principle of module design,we develop a data management module and an intelligent diagnosis module,respectively to achieve database management and intelligent diagnosis function.Specifically,the toolkit used for interface design and other function design is the PyQt application development framework combining Python and Qt.In terms of database design,SQLite and its Python interface are used to build three kinds of user databases.Finally,the system test results show that all the functions of the three user subsystems designed in this paper are realized,and the system as a whole has the characteristics of simplicity and easy operation.Conclusions: First of all,the results of the deep learning segmentation algorithm designed in this paper show that it is effective to add dense skip connections and other changes to the network,and the intelligent diagnosis system can accurately provide clinicians with regions of interest in breast cancer labeling.Secondly,the special convolution block and pre-trained model parameters achieved excellent results on the classification network,indicating that the intelligent diagnosis system in this paper can assist doctors to predict whether there is axillary lymph node metastasis of breast cancer.The above results all prove the significance of applying the deep learning algorithm in this paper to breast cancer ultrasound images.Therefore,through the intelligent diagnosis algorithm and application system designed in this paper,clinicians can obtain highly accurate diagnosis results,which greatly reduces the daily work burden of doctors and improves their work efficiency.As well patients can easily obtain their diagnostic reports in the system;The easy-to-learn operation language and simple database design also reduce the maintenance and management costs of administrators.