Projects

CovEcho

Lung ultrasound (LUS) is possibly the only medical imaging modality that could be used for continuous and periodic monitoring of the lung. This is extremely useful in tracking the lung manifestations either during the onset of lung infection or to track the effect of vaccination on the lung as in pandemics such as COVID-19. There have been many attempts in automating the classification of severity of lung into various classes or automatic segmentation of various LUS landmarks and manifestations. However, all these approaches are based on training static machine learning models which require a significantly large clinically annotated dataset and are computationally heavy and most of the time non-real-time.

In this work, a real-time light weight active learning-based approach is presented for faster triaging in COVID-19 subjects in resource-constrained settings. The tool, based on you look only once (YOLO) network, has the capability of providing the quality of images based on the identification of various LUS landmarks, artifacts and manifestations, prediction of severity of lung infection, the possibility of active learning based on the feedback from clinicians or on the image quality and a summarization of the significant frames which are having high severity of the infection and high image quality for further analysis. The results show that the proposed tool has a mean average precision (mAP) of 66% at an Intersection over Union (IoU) threshold of 0.5 for the prediction of LUS landmarks with initial training on less than 1000 images. The 14MB lightweight YOLOv5s network achieves 123 FPS while running on a Quadro P4000 GPU. The tool is available for usage and analysis upon request from the authors and details can be found online.

LsD-n

Detection and mapping of landslides is an important procedure, to know the extent of landslide phenomena in a certain region, to investigate the recurrence and statistics of slope failures, and to determine the landslide susceptibility, hazard, vulnerability, and risk. Current techniques by means of satellite, airborne and terrestrial remote sensing, facilitate the production of landslide maps, reducing the time and resources required for their compilation and systematic update. In this work, we have investigated recent and existing technologies for landslide mapping by considering its advantages and limitations and proposed a new methodology for automatic landslide delineation using python open source tools. The new methodology handles the above problem using a deep Convolutional Neural Network (CNN). The recent success in deep learning-based feature identification and classification motivated us to apply the same formula in the remote sensing field too.

The proposed network called LsD-n(Landslide Detection network), consists of two parallel networks, fused together resulting in a simpler model with fewer network parameters. The parallel streams of LsD-n, one is a detector CNN which extracts deep representations of landslide candidates from post landslide image data and the other is a detector MLP, which learns from handcrafted features. The detector MLP layer enhances the deep feature map of detector CNN by concatenation/ fusion of features. The fused LsD-network will predict each pixel as landslide or non-landslide with a soft-max classifier applied onto the concatenated feature maps. The proposed fused feature network methodology achieves ≈20% improvement in landslide detection compared to a simple CNN mapping. The proposed method also helps to improve the quality of landslide maps with very good sensitivity (87% on testing data) and is able to map even small landslides of size 33m2 within minutes.

ViEW

Conventional Colonoscopy (CC), through the insertion of a camera attached flexible tube into the intestine, is considered as a standard procedure for direct visualization of the colon and concurrent removal of potentially cancerous lesions. It is a time consuming method and the greatest discomfort caused by CC makes this procedure undesirable. Virtual Colonoscopy (VC) is an imaging test to visualize the 3D model of colon and navigation of a virtual camera through the same tubular structure, without having to insert any physical device into the body. The purpose of this study is to develop a medical workstation for VC which constitutes better image segmentation as well as interactive navigation techniques. The software is implemented using Microsoft Visual C++ 2013, using MFC (Microsoft Foundation Classes) application framework. The DICOM (Digital Imaging and Communications in Medicine) data consists of images of distended colon and meta information of the polyps present, obtained from TCIA (The Cancer Imaging Archive) collections is used for the study.

Page updated

Google Sites

Report abuse