Research Projects

The present project aims at mitigating this shortcoming so that AI based methods can be used in aiding diagnosis, management and rehabilitation of such patients. Also, the related work in the literature suffers due to lack of data availability, as no 3D volumetric data of traumatic injuries is available online. This will also help in generating a repository of CT scans for further work on this subject. In summary, the aim is a) Instant diagnosis b) Accurate diagnosis c) Remote usage especially in rural centres d) Better patient management e) Better patient triage and prognostication Study Objectives a) Development of segmentation model of hemorrhage based on Deep Learning models b) Identification and characterization of secondary brain injuries like trans compartmental herniation’s etc c) Detection and classification of aneurysms based on deep learning models d) Development, validation and testing of training paradigms for traumatic hemorrhage patients e) Detection and classification of skull fractures using machine-based learning models Novelty/Innovation: a) Traumatic Hemorrhage detection & classification by utilizing Computed Tomography images on both online available and real-time dataset b) Segmentation of different types of Traumatic Hemorrhages like Subdural Hemorrhage (SDH), Epidural Hemorrhage (EDH), Intraparenchymal Hemorrhage (IPH), Intraventricular Hemorrhage (IVH) c) Volumetric analysis and quantification of affected region by Intracranial Hemorrhages (ICH) that will help radiologists in quick diagnosis of growth of hematoma in accidental traumatic cases d) Computer Aided Diagnosis system using modified CNN Architectures for detection and quantification of real-time data from calvarias CT scans to assess for fractures for faster and accurate analysis MODEL-1: Develop CNN Architecture for Classification of Hemorrhages The most popular CNN models such AlexNet, VGG-16, VGG-19, GoogLeNet, Inceptionv3, Resnet50, Resnet101 and InceptionResNetV2, have made substantial advancement in image characterisation as labelled dataset is already present predominantly via Image Net. However, radiological glossed dataset in medical imaging domain is unaccounted. New and effective CNN architectures must be developed for medical image classification. MODEL-2: Develop segmentation model for brain hemorrhage detection The deep learning based model for detection of brain hemorrhage and segmentation of bleed using CT images is being developed and tested on real-time dataset. Moreover texture analysis using image processing techniques are also done for accurate characterisation Hemmorhage, thus leads to better diagnostic and treatment planning. Detection and classification of aneurysms are also done using deep learning based models on CT-angiography 3D volumetric data which helps in surgical planning.