The Design And Implementation Of Aided Diagnostic System For Laryngoscope Vocal Cord Lesions Based On Deep Learning

In recent years,the incidence of vocal nodules and vocal cord polyps has increased,and the images of the two are similar.However,the causes and treatment methods are different.The most important thing is that the cancerous rate of vocal cord polyps has also increased year by year.Therefore,it is great significance to correctly distinguishing the vocal nodules and vocal cord polyps.This paper studies the design and implementation of the auxiliary diagnostic system for vocal cord lesions from two aspects:vocal cord image acquisition and vocal cord image processing.In the aspect of vocal cord image acquisition,we investigated the existing video laryngoscope in the market,designed and implemented the wireless HD video laryngoscope system based on WiFi for its defects.The system improves the video resolution to 640*480,and can be transmitted in real time.The transmission delay control is around 200ms.Wireless transmission can reduce the volume of the laryngoscope and facilitate the operation of doctors.Digital transmission is used to facilitate the storage and processing of image data in the later stage.In the aspect of vocal cord image processing,based on the comparison of traditional shallow learning and deep learning algorithms in CAD applications,we chose the deep learning algorithm to extract features directly from the training data set,reduce the depth of human intervention to the system.And research the development history of deep learning algorithm and its application in the field of medical image.With vocal cord images as research objects,construct a vocal cord lesion aided diagnosis system based on convolutional neural network use the deep learning algorithm and adopted AlexNet model which has unique advantages in medical image processing.Use the data augmentation and transfer learning,solve the conflicts that the number of vocal cord image data sets is small and the training of the deep convolutional neural network model requires many data,make the system can better distinguish normal vocal cords,vocal nodules and vocal cord polyps.