Scoring COVID-19 Anomalies on Android Device


Chest computed tomography (CT) imaging has be-come indispensable for staging and managing of COVID-19, and current evaluation of anomalies/abnormalities associated withCOVID-19 has been performed majorly by visual score. The development of automated methods for quantifying COVID-19abnormalities in these CT images is invaluable to clinicians. The hallmark of COVID-19 in chest CT images is the presence of ground-glass opacities in the lung region, which are tedious to segment manually. We propose anamorphic depth embedding based light-weight CNN, called Anam-Net, to segment anomalies in COVID-19 Chest CT images. The proposed Anam-Net has 7.8times fewer parameters compared to the state-of-the-art UNet (or its variants), making it light-weight capable of providing inferences in mobile or resource constraint (point-of-care) platforms.The results from chest CT images (test cases) across different experiments showed that the proposed method could provide good Dice similarity scores for abnormal as well as normal regions in the lung. We have benchmarked Anam-Net with other state-of-the-art architectures like ENet, LEDNet, UNet++, SegNet, Attention UNet and DeepLabV3+. The proposed Anam-Net was also deployed on embedded systems like Raspberry Pi4, NVIDIA Jetson Xavier, and mobile based Android application (CovSeg) embedded with Anam-Net to demonstrate its suitability for point-of-care platforms.


Tracking by Detection framework for counting and localizing cells in a given video sequence.

​​The count and properties of the cells in human body fluids provide vital information about health and are of key interest in bio-medical research. The recent developing field in the cell imaging and microscopy is Imaging Cytometry (IC). The advantage of IC is that it has a very high throughput of the order of several hundred cells per second, as the (live) cells are imaged. But compared to microscopy the morphological features of the cell do not appear to be discriminative and this makes the counting, characterization and classification much more challenging. However, as each instance of a cell is imaged, we propose that efficient tracking of the cells play a key role in enhanced feature extraction and classification in this setting. This further provides additional and complimentary information about the motion characteristics of the cells for better diagnosis. This work aims at automating the process of detecting and tracking Red Blood Cells (RBCs) in IC. We specifically, propose an extension of efficient and fast Kernelized Correlation Filter (KCF) referred to as "Discriminative Correlation Filter Tracking by Blob Detection'' for multi-object tracking in this scenario along with an appropriate association mechanism for efficient tracking of RBCs. In spite of their shape deformations and several crossovers of similar looking cells in poor contrast environments, our approach gives an overall tracking accuracy of 92%.

Media1.avi

Tracking RBCs in Imaging Cytometry


EE5101 IIT T : Detection and Counting of Red blood cells.


The count and properties of the cells in human body fluids provide vital information about health and are of key interest in biomedical research. The recent developing field in cell imaging and microscopy is Imaging Cytometry (IC). The advantage of IC is that it has a very high throughput of the order of several hundred cells per second, as the (live) cells are imaged. But compared to microscopy the morphological features of the cell do not appear to be discriminative and this makes the counting, characterization and classification much more challenging.This project aims at developing an automated algorithm for the detection of RBCs in a given video sequence using basic image processing techniques that are based on region properties.

Media2.avi

E9205 IISc : Scale Correction for Visual Tracking.


Visual tracking is one of the most challenging problems in computer vision. Given the object of interest in the first frame, the role of the tracker is to follow the trajectory of the target. However, there are several challenges such as scale variation, illumination changes, blurring effects, and occlusion which are yet to be tackled with a sufficient degree of realism. In this project, we addressed scale variation by learning a scale parameter that resizes the bounding box (bbox) prediction given by the tracker. The project report can be found here

Media3.avi