UNet Deep Learning Architecture for Segmentation of Vascular and Non-Vascular Images: A Microscopic Look at UNet Components Buffered With Pruning, Explainable Artificial Intelligence, and Bias

Biomedical image segmentation (BIS) task is challenging due to the variations in organ types, position, shape, size, scale, orientation, and image contrast. Conventional methods lack accurate and automated designs. Artificial intelligence (AI)-based UNet has recently dominated BIS. This is the first review of its kind that microscopically addressed UNet types by complexity, stratification of UNet by its components, addressing UNet in vascular vs. non-vascular framework, the key to segmentation challenge vs. UNet-based architecture, and finally interfacing the three facets of AI, the pruning, the explainable AI (XAI), and the AI-bias. PRISMA was used to select 267 UNet-based studies. Five classes were identified and labeled as conventional UNet, superior UNet, attention-channel UNet, hybrid UNet, and ensemble UNet. We discovered 81 variations of UNet by considering six kinds of components, namely encoder, decoder, skip connection, bridge network, loss function, and their combination. Vascular vs. non-vascular UNet architecture was compared. AP(ai)Bias 2.0-UNet was identified in these UNet classes based on (i) attributes of UNet architecture and its performance, (ii) explainable AI (XAI), and, (iii) pruning (compression). Five bias methods such as (i) ranking, (ii) radial, (iii) regional area, (iv) PROBAST, and (v) ROBINS-I were applied and compared using a Venn diagram. Vascular and non-vascular UNet systems dominated with sUNet classes with attention. Most of the studies suffered from a low interest in XAI and pruning strategies. None of the UNet models qualified to be bias-free. There is a need to move from paper-to-practice paradigms for clinical evaluation and settings.

A Circuit-embedded Reservoir Computer for Smart Noise Reduction of MCG Signals

With the COVID-19 pandemic, it has become necessary to monitor cardiac activities, not only for heart patients but for everyone. However, the traditional way to use heavy machines which are non-portable, intrusive, to check the electrocardiography (ECG) is not possible for everyone. As an alternative, there are sensors that can collect magnetocardiography (MCG) signals by measuring the magnetic field produced by the electrical currents in the heart and can be converted into ECG signals. The sensor for MCG is very sensitive, consume low power, portable, and can be a good alternative to check cardiac activities. But the challenging part of these sensors would be the noise at the low frequencies because the heart also oscillates at the low frequencies. As the relevant signal and noise share the same spectral properties, standard linear filtering techniques are not efficient. In this paper, we propose a physical reservoir computing technique using a circuit that can act as a reservoir and a lightweight machine learning model. The output is modeled to reduce the noise and extract the ECG signals out of the MCG ones.

A Hybrid AI Model for Improving COVID-19 Sentiment Analysis in Social Networks

The recent COVID-19 (novel coronavirus disease) pandemic induced a deep polarization among regional as well as global communities. The sentiments regarding the pandemic and its impact on lifestyle and economy, often expressed via social networks, are regarded as critical metrics for capturing such polarization and formulating appropriate intervention by the relevant authorities. While there exist a myriad of Natural Language Processing (NLP) models for mining social media data, we demonstrate the shortcomings of the individual models in this paper, and explore how to improve the COVID-19 sentiment analysis in social media network data via two hybrid predictive models based on a Long-Short-Term-Memory (LSTM)-based autoencoder and a Convolutional Neural Network (CNN) model coupled with a bi-directional LSTM. Through extensive experiments on the recently acquired Twitter dataset, we compare the COVID-19 sentiments exhibited in the USA and Canada using our proposed hybrid predictive models and demonstrate their superiority over individual Artificial Intelligence (AI) models. © 2022 IEEE.

Clinical characteristics of COVID-19 infected cancer patients: A retrospective study in Royal Wolverhampton NHS Trust, United Kingdom

INTRODUCTION: COVID-19 has been declared as a pandemic by the World Health Organisation (WHO)in December 2019, as it spread globally and confirmed cases approach 5,000 000 patients and will exceed 365000 deaths on the 25 May 2020 across over 160 countries. Cancer patients are one of the most vulnerable groups in the current (COVID-19) pandemic. To date, the clinical characteristics of COVID-19-infected cancer patients remain widely not well understood. Patients and methods A retrospective study was conducted in Royal Wolverhampton NHS Trust for COVID-19 Cancer patients. Hospitalised cancer patients diagnosed with COVID-19 infection were identified between 30th March 2020 to 30th June 2020. Patients already have been diagnosed with cancer and had a laboratory-confirmed SARS-CoV-2 infection were enrolled. Clinical retrospective data were collected from hospital medical records, including demographic features, clinical features, laboratory findings, and chest radiograph and chest computed tomography (CT) images. Statistical analysis was done to assess the risk factors associated with severe events which required admission to an intensive care unit, the use of mechanical ventilation, or death Results Forty Cancer patients with Covid 19 infection during the period from 30th March 2020 to 30th June 2020 were enrolled. (52.6%) 22 of patients were females. Median age was 65 years .All patients were local residents of Wolverhampton. Among the cancer patients, Breast cancer was the most frequent type of cancer (n= 9;21.1%), followed by Gl cancers (n= 8;21%) and lymphoma (n = 6;15.8%).Twenty two patients (52.6%) were diagnosed with stage I-III cancer.18 patients (47.4%) were on active chemotherapy, 3 patients were on target therapy and 3 patients(7.9%) were on active immunotherapy. In addition to cancer, 31 (81.6%) patients had at least one or more coexisting chronic diseases. The most common clinical features on admission were fever (92.1%), dry cough (86.8%), and fatigue (92%);29 (76.3%) patients developed dyspnoea along with lymphopaenia (n = 32, 84.2%), high level C-reactive protein (n = 40, 100%), anaemia (n = 22, 57.9 %), and hypoproteinaemia (n = 21, 55.3%). The common chest computed tomography (CT) findings were ground-glass opacity (n = 13) and patchy consolidation (n= 4) .It is important to note that CT chest not done in 17 patients. A total of 19 patients had severe events and the mortality rate was (44.7%) .Median days of hospital admission was (12.5).It is noted that all patients with active immunotherapy had recovered despite disease progression. Conclusions: Cancer patients have deteriorating conditions and worse outcomes from the COVID-19 infection. It is recommended that cancer patients receiving antitumour therapies should have regular screening for COVID-19 infection and should avoid treatments causing immunosuppression or have dose reduction during COVID-19 Pandemic in second wave .Covid 19 has different response with patients on active immunotherapy need to be highlighted.

An Explainable AI Model for Interpretable Lung Disease Classification

In this paper, we develop a framework for lung disease identification from chest X-ray images by differentiating the novel coronavirus disease (COVID-19) or other disease-induced lung opacity samples from normal cases. We perform image processing tasks, segmentation, and train a customized Convolutional Neural Network (CNN) that obtains reasonable performance in terms of classification accuracy. To address the black-box nature of this complex classification model, which emerged as a key barrier to applying such Artificial Intelligence (AI)-based methods for automating medical decisions raising skepticism among clinicians, we address the need to quantitatively interpret the performance of our adopted approach using a Layer-wise Relevance Propagation (LRP)-based method. We also used a pixel flipping-based, robust performance metric to evaluate the explainability of our adopted LRP method and compare its performance with other explainable methods, such as Local Interpretable Model Agnostic Explanation (LIME), Guided Backpropagation (GB), and Deep Taylor Decomposition (DTD). © 2021 IEEE.

On COVID-19 Prediction Using Asynchronous Federated Learning-Based Agile Radiograph Screening Booths

To combat the novel coronavirus (COVID-19) spread, the adoption of technologies including the Internet of Things (IoT) and deep learning is on the rise. However, the seamless integration of IoT devices and deep learning models for radiograph detection to identify the presence of glass opacities and other features in the lung is yet to be envisioned. Moreover, the privacy issue of the collected radiograph data and other health data of the patients has also arisen much concern. To address these challenges, in this paper, we envision a federated learning model for COVID-19 prediction from radiograph images acquired by an X-ray device within a mobile and deployable screening resource booth node (RBN). Our envisioned model permits the privacy-preservation of the acquired radiograph by performing localized learning. We further customize the proposed federated learning model by asynchronously updating the shallow and deep model parameters so that precious communication bandwidth can be spared. Based on a real dataset, the effectiveness of our envisioned approach is demonstrated and compared with baseline methods.

LiHEA: Migrating EEG Analytics to Ultra-Edge IoT Devices With Logic-in-Headbands

Traditional cloud computing of raw Electroencephalogram (EEG) data, particularly for continuous monitoring use-cases, consumes precious network resources and contributes to delay. Motivated by the paradigm shift of edge computing and Internet of Things (IoT) for continuous monitoring, we focus on this paper on the first step to carry out EEG edge analytics at the last frontier (i.e., the ultra-edge) of our considered cyber-physical system for ensuring users' convenience and privacy. To overcome challenges due to computational and energy resource constraints of IoT devices (e.g., EEG headbands/headsets), in this paper, we envision a smart, lightweight model, referred to as Logic-in-Headbands based Edge Analytics (LiHEA), which can be seamlessly incorporated with the consumer-grade EEG headsets to reduce delay and bandwidth consumption. By systematically investigating various traditional machine and deep learning models, we identify and select the best model for our envisioned LiHEA. We consider a use-case for detecting confusion, representing levels of distraction, during online course delivery which has become pervasive during the novel coronavirus (COVID-19) pandemic. We apply a unique feature selection technique to find out which features are triggered with confusion where delta waves, attention, and theta waves were announced as the three most important features. Among various traditional machine and deep learning models, our customized random forest model demonstrated the highest accuracy of 90%. Since the dataset size might have impacted the performance of deep learning-based approaches, we further apply the deep convolutional generative adversarial network (DCGAN) to generate synthetic traces with representative samples of the original EEG data, and thereby enhance the variation in the data. While the performances of the deep learning models significantly increase after the data augmentation, they still cannot outperform the random forest model. Furthermore, computational complexity analysis is performed for the three best-performing algorithms, and random forest emerges as the most viable model for our envisioned LiHEA.

CDK4/6 inhibitors in clinical practice during COVID-19 pandemic: Comparative analysis of the clinical impact of the pandemic for metastatic breast cancer patients during first and second waves in the UK

Background: Nowadays, the (SARS-CoV-2 or COVID19) is an ongoing worldwide pandemic. Substantial changes in the management of metastatic breast cancer patients have been required worldwide in response to the COVID-19 pandemic. This study is focused to evaluate the adoption of the new guidelines for MBC patients who had been treated with CDK4/6 inhibitors. Methods: We had 2 groups of Metastatic BC ER+ HER2- patients who started the CDK4/6 inhibitors during 2 periods, from 1st June 2019 to 30th June 2020 and from 30th June 2020 to 31th October 2020, this includes 45 and 42 patients respectively. The international ESMO COVID-19 & national (NICE) guidelines were implemented. All changes to treatment were conducted in view of delay, omission or reduction of the dose to assess what were the most frequent implications of these changes. Toxicities assessed based on CTCAE 4.0. Data was compared between first and second wave of the pandemic in the UK. The evidence reported reflects the experience matured at our Trust. Results: Two groups of total of 87 patients were enrolled in the study who started treatment with CDK4/6 inhibitors during both waves of the COVID-19 pandemic. Comparison of the 2 groups revealed that 95.3% of patients were females. 45 and 42 patients had Palbociclib and abemaciclib. 13 of them aged > 75 years old. In the first wave: 33.3% had dose reduction. 8 Patients had progression. 15.6% had toxicity required admission. 57.8% continued the treatment without significant toxicities. In the second wave: 52.4% had dose reduction, of those 38.1% due to toxicities. DP in 4 patients. 23.8% had toxicity admission. 88.1% continued the treatment with non significant toxicities. The mortality was 4% in second group in comparison to 11.9% in the first one. No Covid-19 related death reported. Conclusions: The data comparing the oncological outcomes in patients who had their treatment during both waves of the pandemic confirmed the safety of the delivery of the CDK4/6 inhibitors during COVID-19. The Dose reduction has led to more tolerability and does not affect the efficacy. The neutropenia associated with CDK4/6 inhibitors is unlikely to increase the risk of Covid-19 infection. Legal entity responsible for the study: The authors. Funding: Has not received any funding. Disclosure: All authors have declared no conflicts of interest.

UV-CDS: An Energy-Efficient Scheduling of UAVs for Premises Sterilization

Despite the severity of the second wave of the novel coronavirus disease (COVID-19) and the recent hope for vaccine roll-outs, many public and private institutions are forced to resume their activities subject to ensuring an adequate sterilization of their premises. The existing off-the-shelf drones for such environment sanitization have limited flight-time and payload-carrying capacity. In this paper, we address this challenge by formulating an optimization problem to minimize the energy consumed by drones equipped with ultraviolet-C band (UV-C) panels. To solve this computationally hard problem, we propose several heuristics, such as a randomized path selection algorithm whose solution is further improved with a genetic algorithm-based UV-C drone-based sterilization (UV-CDS) scheduling technique. We consider educational institutions, confronting increasing infections, as an important use-case for the problem. Due to the energy constraint of the drones, the number of drones required for sterilization of the campus is smartly altered for various campus scenarios. The respective energy-efficient paths in the proposed heuristics and our envisioned UV-CDS are estimated for the drones. The performance is evaluated through extensive computer-based simulations which clearly demonstrates the effectiveness of UV-CDS in terms of sub-optimal performance and much faster execution time in contrast with the other methods. IEEE