A Deep Learning Framework With Optimizations For Automatic Detection And Localization Of Dendritic Spine

Abstract

With the emergence of Artificial Intelligence (AI), various problems in healthcare industry are being solved. Dendritic spines are protrusions that occur on dendrites of neurons reflecting indications pertaining to brain functionality. Therefore, dendritic spine detection research assumes significance in healthcare domain. There are many existing efforts towards detecting the dendritic spines from medical images automatically. However, there is need for further optimization in detection process besides localization of dendritic spines. Towards this end, in this paper, we proposed a deep learning based framework which exploits multiple models for efficient detection of dendritic spines. The framework exploits VGG16 model for extracting features from given medical image. The features are further used by faster RCNN model which is the actual dendritic spine detection model. The faster RCNN model exploits region proposal network which could provide extracted region proposals that make the detection process easier and efficient. We proposed an algorithm known as Learning based Dendritic Spine Detection (LbDSD) which exploits deep learning models for efficient detection and localization of dendritic spines. Our empirical study with a benchmark dataset revealed that the proposed deep learning framework and underlying algorithm outperforms existing deep learning based methods with highest accuracy 94.87%.