Research On Few-shot Classification Methods For Biomedical Data

Machine learning especially deep learning technologies promote the development of modern medicine and play an important role in the diagnosis of many diseases.Traditional machine learning methods require a large amount of data to train models.In the field of Smart Health,because the source of data involves patient privacy,it is difficult to obtain massive biomedical data for public research,which brings many challenges to the application of machine learning techniques in this field.This paper mainly focuses on the automatic diagnosis of two types of biomedical data,EEG epilepsy signals and COVID-19 lung CT images,and conducts research on few-shot classification methods for biomedical data.The main work is as follows:(1)Considering the problem of epilepsy diagnosis based on EEG signals,an automatic EEG epilepsy detection algorithm based on multi-view TSK model is proposed.The algorithm uses a multi-view framework to collaboratively process feature data extracted from different views.And in the algorithm,a view-weighted mechanism based on quadratic regularization is proposed to distinguish the importance of each view,and it can increase the inner relationship of EEG small dataset from different dimensions.Therefore,we can avoid the performance degradation problem caused by a single feature extraction method.In addition,the algorithm takes the TSK fuzzy system as the basic model,which improves the interpretability of the algorithm.Our experimental results show that,compared with other epilepsy EEG signal detection methods,the method proposed in this paper can obtain more valuable feature information from the data,which can enhance the generalization ability of the model and improve the classification accuracy to have a better performance.Moreover,the method proposed in this paper has better interpretability for the results.(2)Considering the diagnosis of COVID-19 based on lung CT images,an automatic detection algorithm for COVID-19 CT images is proposed.The algorithm uses a deep generative model VAE to increase the scale of the COVID-19 dataset,thereby providing more sample information to reduce overfitting.And based on transfer learning technology,a classification model for COVID-19 diagnosis is proposed.In addition,in order to improve the generalization ability of the model,the automatic diagnosis model is homogeneously integrated using the Stacking ensemble technique.The experimental results on the COVID-19 CT dataset show that,compared with other algorithms,the algorithm proposed in this paper can obtain more valuable sample information from the data,effectively reduce the overfitting of the model,and achieve higher recognition accuracy for CT images of COVID-19.(3)On the basis of the automatic COVID-19 detection algorithm proposed in this paper,an intelligent COVID-19 detection system based on CT images is developed.We make some analysis about the system to determine the various functional modules and system architecture,and then use Python language and PyQt framework to develop the system agile.The core of the COVID-19 detection system is the COVID-19 detection algorithm proposed in this paper.By combining the algorithm and the system,the rapid and convenient diagnosis for COVID-19 is realized,which reduces the workload of doctors.