Deep Spiking Convolutional Neural Networks For Tuberculosis Diagnosis

Tuberculosis is a common pulmonary infectious disease caused by Mycobacterium tuberculosis,which seriously endangers human life and health.With the rapid development of artificial intelligence technology,the disease auxiliary diagnosis system based on artificial intelligence has been widely used in large hospitals at home and abroad.Artificial intelligence technology has become a major means of chest X-ray and other medical image processing.Therefore,it is necessary to study the auxiliary diagnosis of tuberculosis disease based on artificial intelligence technology,which is of great significance in reducing the diagnostic pressure of doctors and improving the emergency response ability of hospitals.Deep convolution neural network has powerful image pattern classification ability and has become the preferred solution for image recognition and classification,target detection and other computer vision tasks.The spiking neural network takes into account the changes of membrane potential and the process of spike firing of biological neurons and uses accurate spike timing coding mode for information transmission and processing,which can more closely describe the actual biological neural system,so as to achieve efficient information processing.In order to realize powerful artificial intelligence,more complex and refined brain-inspired computing models are needed.In this study,a deep spiking convolution neural network model is proposed and applied to the auxiliary diagnosis of pulmonary tuberculosis.The main contents are as follows:(1)Combining the deep convolution neural network with powerful image classification ability and the spiking neural network with biological plausibility,a pulmonary tuberculosis image classification model based on deep spiking convolution neural network is proposed.When training the model,the combination of unsupervised learning based on spike-timing-dependent plasticity rule and reinforcement learning based on reward modulated spike-timing-dependent plasticity is used to realize network synaptic weight learning.The proposed model is applied to Montgomery chest X-ray data set for tuberculosis diagnosis to verify the classification performance of the model.At the same time,the influence of some important parameters of the model on the classification performance of the model is analyzed.Experimental results show that the proposed model can accurately classify the input chest X-ray images.(2)An auxiliary diagnosis system for pulmonary tuberculosis was designed and developed by using the proposed deep spiking convolution neural network model.The system can give the prediction results of tuberculosis disease according to the chest Xray of the medical record,and the user needs to confirm the prediction results.The distinguishing feature of the system is that it can make corrections based on the confirmed diagnosis opinions and retrain the prediction model to continuously improve the prediction accuracy of the model and enhance the effect of tuberculosis diagnosis.The system is easy to operate,and it can help doctors in the auxiliary diagnosis of pulmonary tuberculosis,and has an established practical value.