Identification of Neural Correlates of Face Recognition Using Machine Learning Approach

Object recognition has always been one of the key areas of research in modern times, especially in healthcare and engineering industry, because of the wide range of applications it has. There have been various methods to recognize and classify objects like shape matching, color matching, sliding window approach, etc. but a common problem the computational models face is the appropriate representation of 3D objects, image variation with angle variation, illumination effects, and the high computational costs of these models to maintain output accuracy. In this paper, we propose a novel method to detect and analyze the process of object recognition from magnetoencephalogram (MEG) signals of human brain. This could be made possible by classifying the object as face/scrambled face using machine learning through support vector machine (SVM). We train our SVM model using the recordings in DecMeg Human Brain dataset obtained from Kaggle. In addition, by calculating the accuracy of individual sensors for the duration, we are able to identify the cluster of sensors responsible for visual recognition and the dynamic interaction among sensors with the passage of time using neural coordinates of the magnetometer sensors with an accuracy of 74.85%. We found that the sensors in the occipitotemporal and occipitoparietal lobes are most actively involved in visual classification. The proposed approach has been able to reduce the previous effective time stamp of 100–360 ms to 124–240 ms. This reduces the computational cost of the model while establishing the essential relationships between MEG signals and facial detection.