Research On Medical Image Segmentation And Classification Methods Based On Deep Neural Networks

The use of deep learning methods to process medical images can help improve the efficiency of clinical diagnosis.In this paper,we conducted specific research on fetal head ultrasound image segmentation and Coronavirus Disease 2019(COVID-19)chest CT image classification.Fetal head circumference is one of the most important biological characteristics for evaluating fetal development,but manual measurement has operator errors,and is time-consuming and labor-intensive.The innovations of this paper are as follows :(1)Based on the thought of joint training of the deep neural network Mask R-CNN,and according to the characteristics of the fetal head close to ellipse shape in the ultrasound image,a branch of the head circumference measurement loss function was proposed;(2)After the segmentation branch of the original model,the Elli Fit algorithm was used to fit the ellipse of the segmentation mask,and the perimeter of the fitted ellipse was calculated using the Ramanujan formula as the head circumference measurement value.The true value of the head circumference and the mean square error of the measurement value were taken as the head circumference measurement loss function.The function was added to the original loss function,which made the training process of the model closely related to the measurement task.190 fetal head ultrasound images were tested,the Dice coefficient was(96.89% ± 1.01)%,the measurement error was(0.33± 1.54)mm,and the average processing time for an ultrasound image was 0.33 s.Compared with the traditional manual measurement method or the original model,the proposed method improved the speed by 1.13?16.87 s and the accuracy by 0.21?1.68 mm.The results showed that the improved Mask R-CNN could help for the efficiency of doctors in measuring fetal head circumference,and could meet clinical needs.COVID-19 has caused hundreds of thousands of infections and deaths.Efficient diagnostic methods could help curb its global spread.The purpose of this study was to develop and evaluate a method for accurately diagnosing COVID-19 based on computed tomography scans in real time.We proposed an architecture named concatenated feature pyramid network(Concat-FPN)with an attention mechanism,by concatenating feature maps of multiple.The proposed architecture was then used to form COVID-CT-GAN and COVID-CT-Dense Net.The former for data augmentation,and the latter for data classification.The proposed method was evaluated on 3 different numbers of magnitude of COVID-19 computed tomography datasets.COVID-CT-GAN increased the accuracy by 2% to 3%,the recall by 2% to 4%,the precision by 1% to 3%,the F1-score by 1% to 3%,and the area under the curve by 1% to 4%.COVID-CT-Dense Net increased the accuracy by 1% to 3%,the recall by 4% to 9%,the precision by 1%,the F1-score by 1% to 3%,and the area under the curve by 2%.The experimental results showed that our method improves the efficiency of diagnosing COVID-19 on computed tomography images,and helped overcome the problem of limited training data when using deep learning methods to diagnose COVID-19.