Alex Ter-SarkisovCity, University of London · Department of Computer Science
Alex Ter-Sarkisov
PhD
About
40
Publications
8,969
Reads
How we measure 'reads'
A 'read' is counted each time someone views a publication summary (such as the title, abstract, and list of authors), clicks on a figure, or views or downloads the full-text. Learn more
261
Citations
Citations since 2017
Introduction
COVID-19 Prediction + Lesion Detection/Segmentation From Chest CT Scans
Skills and Expertise
Publications
Publications (40)
We present COVID-CT-Mask-Net model that predicts COVID-19 from CT scans. The model works in two stages: first, it detects the instances of ground glass opacity and consolidation in CT scans, then predicts the condition from the ranked bounding box detections. To develop the solution for the three-class problem (COVID, common pneumonia and control),...
We introduce a lightweight model based on Mask R-CNN with ResNet18 and ResNet34 backbone models that segments lesions and predicts COVID-19 from chest CT scans in a single shot. The model requires a small dataset to train: 650 images for the segmentation branch and 3000 for the classification branch, and it is evaluated on 21292 images to achieve a...
We present a model that fuses instance segmentation, Long Short-Term Memory Network and Attention mechanism to predict COVID-19 and segment chest CT scans. The model works by extracting a sequence of Regions of Interest that contain class-relevant information, and applies two Long Short-Term Memory networks with attention to this sequence to extrac...
We present COVID-CT-Mask-Net model that predicts COVID-19 in chest CT scans. The model works in two stages: in the first stage, Mask R-CNN is trained to localize and detect two types of lesions in images. In the second stage, these detections are fused to classify the whole input image. To develop the solution for the three-class problem (COVID-19,...
In this paper we introduce Local Logo Generative Adversarial Network (LL-GAN) that uses regional features extracted from Faster R-CNN for logo generation. We demonstrate the strength of this approach by training the framework on a small style-rich dataset of real heavy metal logos to generate new ones. LL-GAN achieves Inception Score of 5.29 and Fr...
We present a model that fuses lesion segmentation with Attention Mechanism to predict COVID-19 from chest CT scans. The model segments lesions, extracts Regions of Interest from scans and applies Attention to them to determine the most relevant ones for image classification. Additionally, we augment the model with Long-Short Term Memory Network lay...
We present a novel framework that integrates segmentation of lesion masks and prediction of COVID-19 in chest CT scans in one shot. In order to classify the whole input image, we introduce a type of associations among lesion mask features extracted from the scan slice that we refer to as affinities. First, we map mask features to the affinity space...
In this paper we introduce the Local Logo Generative Adversarial Network (LL-GAN) that uses regional features extracted from the Faster Regional Convolutional Neural Network (Faster R-CNN) to generate logos. We demonstrate the strength of this approach by training the framework on a small style-rich dataset collected online to generate large impres...
This paper investigates the classification of radiographic images with eleven convolutional neural network (CNN) architectures (GoogleNet, VGG-19, AlexNet, SqueezeNet, ResNet-18, Inception-v3, ResNet-50, VGG-16, ResNet-101, DenseNet-201 and Inception-ResNet-v2). The CNNs were used to classify a series of wrist radiographs from the Stanford Musculos...
This paper investigates the classification of radiographic images with eleven convolutional neural network (CNN) architectures (GoogleNet, VGG-19, AlexNet, SqueezeNet, ResNet-18, Inception-v3, ResNet-50, VGG-16, ResNet-101, DenseNet-201 and Inception-ResNet-v2). The CNNs were used to classify a series of wrist radiographs from the Stanford Musculos...
We present a model that fuses instance segmentation, Long Short-Term Memory Network and Attention mechanism to predict COVID-19 and segment chest CT scans. The model works by extracting a sequence of Regions of Interest that contain class-relevant information, and applies two Long Short-Term Memory networks with attention to this sequence to extrac...
![Summary of the CNCB-NCOV [ZLS + 20] classification dataset and...](publication/348229266/figure/tbl1/AS:976733830385679@1609882900694/Summary-of-the-CNCB-NCOV-ZLS-20-classification-dataset-and-COVIDx-CT-splits-GWW20_Q320.jpg)
We introduce a model that segments lesions and predicts COVID-19 from chest CT scans through the derivation of an affinity matrix between lesion masks. The novelty of the methodology is based on the computation of the affinity between the lesion masks' features extracted from the image. First, a batch of vectorized lesion masks is constructed. Then...
We introduce a lightweight model based on Mask R-CNN with ResNet18 and ResNet34 backbone models that segments lesions and predicts COVID-19 from chest CT scans in a single shot. The model requires a small dataset to train: 650 images for the segmentation branch and 3000 for the classification branch, and it is evaluated on 21292 images to achieve a...
In this paper we compare the models for the detection and segmentation of Ground Glass Opacity and Consolidation in chest CT scans. These lesion areas are often associated both with common pneumonia and COVID-19. We train a Mask R-CNN model to segment these areas with high accuracy using three approaches: merging masks for these lesions into one, d...
We introduce a lightweight Mask R-CNN model that segments areas with the Ground Glass Opacity and Consolidation in chest CT scans. The model uses truncated ResNet18 and ResNet34 nets with a single layer of Feature Pyramid Network as a backbone net, thus substantially reducing the number of the parameters and the training time compared to similar so...
In this paper we compare the models for the detection and segmentation of Ground Glass Opacity and Consolidation in chest CT scans. These lesion areas are often associated both with common pneumonia and COVID-19. We train a Mask R-CNN model to segment these areas with high accuracy using three approaches: merging masks for these lesions into one, d...
We introduce a lightweight Mask R-CNN model that segments areas with the Ground Glass Opacity and Consolidation in chest CT scans. The model uses truncated ResNet18 and ResNet34 nets with a single layer of Feature Pyramid Network as a backbone net, thus substantially reducing the number of the parameters and the training time compared to similar so...
We present COVID-CT-Mask-Net model that predicts COVID-19 from CT scans. The model works in two stages: first, it detects the instances of ground glass opacity and consolidation in CT scans, then predicts the condition from the ranked bounding box detections. To develop the solution for the three-class problem (COVID, common pneumonia and control),...
We introduce a lightweight Mask R-CNN model that segments areas with the Ground Glass Opacity and Consolidation in chest CT scans. The model uses truncated ResNet18 and ResNet34 nets with a single layer of Feature Pyramid Network as a backbone net, thus substantially reducing the number of the parameters and the training time compared to similar so...
In this paper we compare the models for the detection and segmentation of Ground Glass Opacity and Consolidation in chest CT scans. These lesion areas are often associated both with common pneumonia and COVID-19. We train a Mask R-CNN model to segment these areas with high accuracy using three approaches: merging masks for these lesions into one, d...
We introduce a method for transferring style from the logos of heavy metal bands onto corporate logos using a VGG16 network. We establish the contribution of different layers and loss coefficients to the learning of style, minimization of artefacts and maintenance of readability of corporate logos. We find layers and loss coefficients that produce...
We introduce a method for transferring style from the logos of heavy metal bands onto corporate logos using a VGG16 network. We establish the contribution of different layers and loss coefficients to the learning of style, minimization of artefacts and maintenance of readability of corporate logos. We find layers and loss coefficients that produce...
We present an instance segmentation algorithm trained and applied to a CCTV recording of beef cattle during a winter finishing period. A fully convolutional network was transformed into an instance segmentation network that learns to label each instance of an animal separately. We introduce a conceptually simple framework that the network uses to o...
In this paper we present a novel instance segmentation algorithm that extends a fully convolutional network to learn to label objects separately without prediction of regions of interest. We trained the new algorithm on a challenging CCTV recording of beef cattle, as well as benchmark MS COCO and Pascal VOC datasets. Extensive experimentation showe...
![α\documentclass[12pt]{minimal} \usepackage{amsmath}...](profile/Alex-Ter-Sarkisov/publication/301560655/figure/fig3/AS:960250052104228@1605952861544/adocumentclass12ptminimal-usepackageamsmath-usepackagewasysym_Q320.jpg)
There has been a variety of crossover operators proposed for Real-Coded Genetic Algorithms (RCGAs), which recombine values from the same location in pairs of strings. In this article we present a recombination operator for RC- GAs that selects the locations randomly in both parents, and compare it to mainstream crossover operators in a set of exper...
This paper introduces a new approach to the long-term tracking of an object in a challenging environment. The object is a cow and the environment is an enclosure in a cowshed. Some of the key challenges in this domain are a cluttered background, low contrast and high similarity between moving objects – which greatly reduces the efficiency of most e...
It is today acknowledged that neural net-
work language models outperform back-
off language models in applications like
speech recognition or statistical machine
translation. However, training these mod-
els on large amounts of data can take sev-
eral days. We present efficient techniques
to adapt a neural network language model
to new data. Inste...
It is today acknowledged that neural network language models outperform back-
off language models in applications like speech recognition or statistical
machine translation. However, training these models on large amounts of data
can take several days. We present efficient techniques to adapt a neural
network language model to new data. Instead of...
In this article a tool for the analysis of population-based EAs is used to
derive asymptotic upper bounds on the optimization time of the algorithm
solving Royal Roads problem, a test function with plateaus of fitness. In
addition to this, limiting distribution of a certain subset of the population
is approximated.
In this article we present an Elitism Levels Traverse Mechanism that we
designed to find bounds on population-based Evolutionary algorithms solving
unimodal functions. We prove its efficiency theoretically and test it on OneMax
function deriving bounds c{\mu}n log n - O({\mu} n). This analysis can be
generalized to any similar algorithm using varia...
We present a number of bounds on convergence time for two elitist
population-based Evolutionary Algorithms using a recombination operator
k-Bit-Swap and a mainstream Randomized Local Search algorithm. We study the
effect of distribution of elite species and population size.
We present an analysis of the performance of an elitist Evolutionary
algorithm using a recombination operator known as 1-Bit-Swap on the Royal Roads
test function based on a population. We derive complete, approximate and
asymptotic convergence rates for the algorithm. The complete model shows the
benefit of the size of the population and re- combi...
We present a number of convergence properties of population-based Evolutionary Algorithms (EAs) on a set of test functions. Focus is on EA using k-Bit-Swap (kBS) operator. We compare our findings to past research.
Genetic algorithms (GA) mostly commonly use three main operators: selection, crossover and mutation, although many others have been proposed in the literature. This article introduces a new operator, k-bit-swap, which swaps bits between two strings without preserving the location of those bits, changing their order of bits in the string. It can be...
Questions
Questions (8)
My paper is under review at one of the OA journals, and my uni may not be able to cover it in full. In addition to the possible fee waiver, etc by the journal, are there external small fundings I could apply for? The paper is about COVID-19 CT scans segmentation and classification.
I'm working with a RVL-CDIP dataset( https://www.cs.cmu.edu/~aharley/rvl-cdip/) with 16 classes labelled at image level (a large collection of documents, ranging from ads to scientific articles). I'm trying to use weights from Faster- and Mask R-CNN pretrained on different ICDAR data (focused scene texts, receipts and similar). The results compared to baseline ResNet18/34/50 models are not very good. In order to pretrain Faster- and Mask R-CNN on a more relevant data, I need a document dataset labelled at object level (boxes or masks), however small. Does such dataset exist?
Just like Mask R-CNN is an extension of Faster R-CNN for instance segmentation, I was looking for a similar extension for YOLO, but couldn't find any. Solutions in pytorch would be especially useful.
In my work on classifying chest CT scans and extending Faster- and Mask R-CNN I found the conversion of batch of regional detections (Faster R-CNN regions of interest) into a feature vector highly useful. Essentially I'm taking a batch of box coordinates (Nx4 dimensions), convert it into a vector size N*4 and use it as an input (i.e. dimensionality 1xN*4) in the classification module.
This is a rather straightforward step when extending local (regional) to global (image-level) predictions. My question is: are there publications that do a similar conversion (batch to features), no matter at what level.
What I did (unpublished yet) is extract several (e.g. 16) mask features within Mask R-CNN from multiple images, construct a distance metric to compare them to predict the image class. Each set of feature vectors is labelled (image label).
The problem is that if I run, e.g. 4 images, that is 4 feedforward+backprops + 16x4=64 vectors, i.e. 64x64 distance metrics, loss computations +backprops. This puts some pressure on GPU VRAM, preventing from using large samples and hence reducing the efficiency. I'd be grateful for an open-source solution of DL+ supervised clustering that allows using large samples from images, if such exists.
