A robust class decomposition-based approach for detecting Alzheimer's progression.

Computer-aided diagnosis of Alzheimer's disease (AD) is a rapidly growing field with the possibility to be utilized in practice. Deep learning has received much attention in detecting AD from structural magnetic resonance imaging (sMRI). However, training a convolutional neural network from scratch is problematic because it requires a lot of annotated data and additional computational time. Transfer learning can offer a promising and practical solution by transferring information learned from other image recognition tasks to medical image classification. Another issue is the dataset distribution's irregularities. A common classification issue in datasets is a class imbalance, where the distribution of samples among the classes is biased. For example, a dataset may contain more instances of some classes than others. Class imbalance is challenging because most machine learning algorithms assume that each class should have an equal number of samples. Models consequently perform poorly in prediction. Class decomposition can address this problem by making learning a dataset's class boundaries easier. Motivated by these approaches, we propose a class decomposition transfer learning (CDTL) approach that employs VGG19, AlexNet, and an entropy-based technique to detect AD from sMRI. This study aims to assess the robustness of the CDTL approach in detecting the cognitive decline of AD using data from various ADNI cohorts to determine whether comparable classification accuracy for the two or more cohorts would be obtained. Furthermore, the proposed model achieved state-of-the-art performance in predicting mild cognitive impairment (MCI)-to-AD conversion with an accuracy of 91.45%.

Explainable hierarchical clustering for patient subtyping and risk prediction.

We present a pipeline in which machine learning techniques are used to automatically identify and evaluate subtypes of hospital patients admitted between 2017 and 2021 in a large UK teaching hospital. Patient clusters are determined using routinely collected hospital data, such as those used in the UK's National Early Warning Score 2 (NEWS2). An iterative, hierarchical clustering process was used to identify the minimum set of relevant features for cluster separation. With the use of state-of-the-art explainability techniques, the identified subtypes are interpreted and assigned clinical meaning, illustrating their robustness. In parallel, clinicians assessed intracluster similarities and intercluster differences of the identified patient subtypes within the context of their clinical knowledge. For each cluster, outcome prediction models were trained and their forecasting ability was illustrated against the NEWS2 of the unclustered patient cohort. These preliminary results suggest that subtype models can outperform the established NEWS2 method, providing improved prediction of patient deterioration. By considering both the computational outputs and clinician-based explanations in patient subtyping, we aim to highlight the mutual benefit of combining machine learning techniques with clinical expertise.

Design and Evaluation of an Intensive Care Unit Dashboard Built in Response to the COVID-19 Pandemic: Semistructured Interview Study.

BACKGROUND: Dashboards and interactive displays are becoming increasingly prevalent in most health care settings and have the potential to streamline access to information, consolidate disparate data sources and deliver new insights. Our research focuses on intensive care units (ICUs) which are heavily instrumented, critical care environments that generate vast amounts of data and frequently require individualized support for each patient. Consequently, clinicians experience a high cognitive load, which can translate to suboptimal performance. The global COVID-19 pandemic exacerbated this problem by generating a large number of additional hospitalizations, which necessitated a new tool that would help manage ICUs' census. In a previous study, we interviewed clinicians at the University Hospitals Bristol and Weston National Health Service Foundation Trust to capture the requirements for bespoke dashboards that would alleviate this problem. OBJECTIVE: This study aims to design, implement, and evaluate an ICU dashboard to allow for monitoring of the high volume of patients in need of critical care, particularly tailored to high-demand situations, such as those seen during the COVID-19 pandemic. METHODS: Building upon the previously gathered requirements, we developed a dashboard, integrated it within the ICU of a National Health Service trust, and allowed all staff to access our tool. For evaluation purposes, participants were recruited and interviewed following a 25-day period during which they were able to use the dashboard clinically. The semistructured interviews followed a topic guide aimed at capturing the usability of the dashboard, supplemented with additional questions asked post hoc to probe themes established during the interview. Interview transcripts were analyzed using a thematic analysis framework that combined inductive and deductive approaches and integrated the Technology Acceptance Model. RESULTS: A total of 10 participants with 4 different roles in the ICU (6 consultants, 2 junior doctors, 1 nurse, and 1 advanced clinical practitioner) participated in the interviews. Our analysis generated 4 key topics that prevailed across the data: our dashboard met the usability requirements of the participants and was found useful and intuitive; participants perceived that it impacted their delivery of patient care by improving the access to the information and better equipping them to do their job; the tool was used in a variety of ways and for different reasons and tasks; and there were barriers to integration of our dashboard into practice, including familiarity with existing systems, which stifled the adoption of our tool. CONCLUSIONS: Our findings show that the perceived utility of the dashboard had a positive impact on the clinicians' workflows in the ICU. Improving access to information translated into more efficient patient care and transformed some of the existing processes. The introduction of our tool was met with positive reception, but its integration during the COVID-19 pandemic limited its adoption into practice.

Disentangling the Link Between Image Statistics and Human Perception.

In the 1950s, Barlow and Attneave hypothesised a link between biological vision and information maximisation. Following Shannon, information was defined using the probability of natural images. A number of physiological and psychophysical phenomena have been derived ever since from principles like info-max, efficient coding, or optimal denoising. However, it remains unclear how this link is expressed in mathematical terms from image probability. First, classical derivations were subjected to strong assumptions on the probability models and on the behaviour of the sensors. Moreover, the direct evaluation of the hypothesis was limited by the inability of the classical image models to deliver accurate estimates of the probability. In this work we directly evaluate image probabilities using an advanced generative model for natural images, and we analyse how probability-related factors can be combined to predict human perception via sensitivity of state-of-the-art subjective image quality metrics. We use information theory and regression analysis to find a combination of just two probability-related factors that achieves 0.8 correlation with subjective metrics. This probability-based sensitivity is psychophysically validated by reproducing the basic trends of the Contrast Sensitivity Function, its suprathreshold variation, and trends of the Weber-law and masking.

A multi-sensor dataset with annotated activities of daily living recorded in a residential setting.

SPHERE is a large multidisciplinary project to research and develop a sensor network to facilitate home healthcare by activity monitoring, specifically towards activities of daily living. It aims to use the latest technologies in low powered sensors, internet of things, machine learning and automated decision making to provide benefits to patients and clinicians. This dataset comprises data collected from a SPHERE sensor network deployment during a set of experiments conducted in the 'SPHERE House' in Bristol, UK, during 2016, including video tracking, accelerometer and environmental sensor data obtained by volunteers undertaking both scripted and non-scripted activities of daily living in a domestic residence. Trained annotators provided ground-truth labels annotating posture, ambulation, activity and location. This dataset is a valuable resource both within and outside the machine learning community, particularly in developing and evaluating algorithms for identifying activities of daily living from multi-modal sensor data in real-world environments. A subset of this dataset was released as a machine learning competition in association with the European Conference on Machine Learning (ECML-PKDD 2016).

A longitudinal observational study of home-based conversations for detecting early dementia: protocol for the CUBOId TV task.

INTRODUCTION: Limitations in effective dementia therapies mean that early diagnosis and monitoring are critical for disease management, but current clinical tools are impractical and/or unreliable, and disregard short-term symptom variability. Behavioural biomarkers of cognitive decline, such as speech, sleep and activity patterns, can manifest prodromal pathological changes. They can be continuously measured at home with smart sensing technologies, and permit leveraging of interpersonal interactions for optimising diagnostic and prognostic performance. Here we describe the ContinUous behavioural Biomarkers Of cognitive Impairment (CUBOId) study, which explores the feasibility of multimodal data fusion for in-home monitoring of mild cognitive impairment (MCI) and early Alzheimer's disease (AD). The report focuses on a subset of CUBOId participants who perform a novel speech task, the 'TV task', designed to track changes in ecologically valid conversations with disease progression. METHODS AND ANALYSIS: CUBOId is a longitudinal observational study. Participants have diagnoses of MCI or AD, and controls are their live-in partners with no such diagnosis. Multimodal activity data were passively acquired from wearables and in-home fixed sensors over timespans of 8-25 months. At two time points participants completed the TV task over 5 days by recording audio of their conversations as they watched a favourite TV programme, with further testing to be completed after removal of the sensor installations. Behavioural testing is supported by neuropsychological assessment for deriving ground truths on cognitive status. Deep learning will be used to generate fused multimodal activity-speech embeddings for optimisation of diagnostic and predictive performance from speech alone. ETHICS AND DISSEMINATION: CUBOId was approved by an NHS Research Ethics Committee (Wales REC; ref: 18/WA/0158) and is sponsored by University of Bristol. It is supported by the National Institute for Health Research Clinical Research Network West of England. Results will be reported at conferences and in peer-reviewed scientific journals.

OPERAnet, a multimodal activity recognition dataset acquired from radio frequency and vision-based sensors.

This paper presents a comprehensive dataset intended to evaluate passive Human Activity Recognition (HAR) and localization techniques with measurements obtained from synchronized Radio-Frequency (RF) devices and vision-based sensors. The dataset consists of RF data including Channel State Information (CSI) extracted from a WiFi Network Interface Card (NIC), Passive WiFi Radar (PWR) built upon a Software Defined Radio (SDR) platform, and Ultra-Wideband (UWB) signals acquired via commercial off-the-shelf hardware. It also consists of vision/Infra-red based data acquired from Kinect sensors. Approximately 8 hours of annotated measurements are provided, which are collected across two rooms from 6 participants performing 6 daily activities. This dataset can be exploited to advance WiFi and vision-based HAR, for example, using pattern recognition, skeletal representation, deep learning algorithms or other novel approaches to accurately recognize human activities. Furthermore, it can potentially be used to passively track a human in an indoor environment. Such datasets are key tools required for the development of new algorithms and methods in the context of smart homes, elderly care, and surveillance applications.

Conditional t-SNE: more informative t-SNE embeddings.

Dimensionality reduction and manifold learning methods such as t-distributed stochastic neighbor embedding (t-SNE) are frequently used to map high-dimensional data into a two-dimensional space to visualize and explore that data. Going beyond the specifics of t-SNE, there are two substantial limitations of any such approach: (1) not all information can be captured in a single two-dimensional embedding, and (2) to well-informed users, the salient structure of such an embedding is often already known, preventing that any real new insights can be obtained. Currently, it is not known how to extract the remaining information in a similarly effective manner. We introduce conditional t-SNE (ct-SNE), a generalization of t-SNE that discounts prior information in the form of labels. This enables obtaining more informative and more relevant embeddings. To achieve this, we propose a conditioned version of the t-SNE objective, obtaining an elegant method with a single integrated objective. We show how to efficiently optimize the objective and study the effects of the extra parameter that ct-SNE has over t-SNE. Qualitative and quantitative empirical results on synthetic and real data show ct-SNE is scalable, effective, and achieves its goal: it allows complementary structure to be captured in the embedding and provided new insights into real data.

Human Activity Recognition Based on Dynamic Active Learning.

Activity of daily living is an important indicator of the health status and functional capabilities of an individual. Activity recognition, which aims at understanding the behavioral patterns of people, has increasingly received attention in recent years. However, there are still a number of challenges confronting the task. First, labelling training data is expensive and time-consuming, leading to limited availability of annotations. Secondly, activities performed by individuals have considerable variability, which renders the generally used supervised learning with a fixed label set unsuitable. To address these issues, we propose a dynamic active learning-based activity recognition method in this work. Different from traditional active learning methods which select samples based on a fixed label set, the proposed method not only selects informative samples from known classes, but also dynamically identifies new activities which are not included in the predefined label set. Starting with a classifier that has access to a limited number of labelled samples, we iteratively extend the training set with informative labels by fully considering the uncertainty, diversity and representativeness of samples, based on which better-informed classifiers can be trained, further reducing the annotation cost. We evaluate the proposed method on two synthetic datasets and two existing benchmark datasets. Experimental results demonstrate that our method not only boosts the activity recognition performance with considerably reduced annotation cost, but also enables adaptive daily activity analysis allowing the presence and detection of novel activities and patterns.

Curation of an intensive care research dataset from routinely collected patient data in an NHS trust.

In this data note we provide the details of a research database of 4831 adult intensive care patients who were treated in the Bristol Royal Infirmary, UK between 2015 and 2019. The purposes of this publication are to describe the dataset for external researchers who may be interested in making use of it, and to detail the methods used to curate the dataset in order to help other intensive care units make secondary use of their routinely collected data. The curation involves linkage between two critical care datasets within our hospital and the accompanying code is available online. For reasons of data privacy the data cannot be shared without researchers obtaining appropriate ethical consents. In the future we hope to obtain a data sharing agreement in order to publicly share the de-identified data, and to link our data with other intensive care units who use a Philips clinical information system.

Towards a decision support tool for intensive care discharge: machine learning algorithm development using electronic healthcare data from MIMIC-III and Bristol, UK.

OBJECTIVE: The primary objective is to develop an automated method for detecting patients that are ready for discharge from intensive care. DESIGN: We used two datasets of routinely collected patient data to test and improve on a set of previously proposed discharge criteria. SETTING: Bristol Royal Infirmary general intensive care unit (GICU). PATIENTS: Two cohorts derived from historical datasets: 1870 intensive care patients from GICU in Bristol, and 7592 from Medical Information Mart for Intensive Care (MIMIC)-III. RESULTS: In both cohorts few successfully discharged patients met all of the discharge criteria. Both a random forest and a logistic classifier, trained using multiple-source cross-validation, demonstrated improved performance over the original criteria and generalised well between the cohorts. The classifiers showed good agreement on which features were most predictive of readiness-for-discharge, and these were generally consistent with clinical experience. By weighting the discharge criteria according to feature importance from the logistic model we showed improved performance over the original criteria, while retaining good interpretability. CONCLUSIONS: Our findings indicate the feasibility of the proposed approach to ready-for-discharge classification, which could complement other risk models of specific adverse outcomes in a future decision support system. Avenues for improvement to produce a clinically useful tool are identified.

A dataset for room level indoor localization using a smart home in a box.

An annotated dataset of measurements obtained using the EurValve Smart Home In a Box (SHIB) rehabilitation monitoring system is presented. The SHiB is a low cost and easily deployable kit designed to collect data from a wrist-worn wearable in a home environment. The data presented is intended to evaluate room level indoor localization methods. The wearable device registers tri-axial accelerometer measurements which are sampled and transmitted as the payload of a Bluetooth Low Energy (BLE) packet. Four receiving gateways, each placed in a different room throughout a typical residential house, extract the accelerometer data and determine a Received Signal Strength Indicator (RSSI) for each received BLE packet. RSSI values can represent propagation losses due to distance or shadowing between the wearable transmitter and the gateway receiver. The dataset is presented in two parts. The first is composed of four calibration or training sequences, carried out by ten participants to offer ground truth calibrations for four rooms in the house. We refer to the calibration phase as the steps taken to gather training data. The calibration procedure was designed to be as straight-forward as possible, to allow a participant to adequately train the SHiB system without supervision. Ten participants each carried out a straight forward calibration procedure once, with four participants carrying out the calibration twice, on different occasions. One participant carried out the calibration on a third occasion. The second part of the data consists of a free-living experiment that was carried out over a period of five and a half hours starting at 7.37 a.m. Of this, one and a half hours of measurements are recorded within a room containing a gateway, where one participant carried out activities of daily living while their ground-truth location was accurately annotated within each room with a gateway present. The calibration data can be used as a training scheme and the living data as a test scenario. The dataset can be found at https://github.com/rymc/a-dataset-for-indoor-localization-using-a-smart-home-in-a-box.

Residential wearable RSSI and accelerometer measurements with detailed location annotations.

This repository offers smart-home wearable accelerometer and Radio Signal Strength Indicator (RSSI) data acquired : 1) with low-cost hardware; 2) with high-resolution location annotations; 3) from four UK homes. The data are intended to evaluate RSSI-based indoor localisation methods with activity measurements provided from a user-worn wearable device. A wrist worn accelerometer records activity signatures which are relayed to a number of receiving Access Points (AP) placed throughout the building. Upon reception of a packet, each AP measures the RSSI of the received radio signal and timestamps the accelerometer measurements. Location labels are recorded automatically using a small camera which registers fiducial floor tags as the participant carries out their normal routines in a natural way. Approximately 14 h of annotated wearable measurements are provided. A scripted fingerprint measurement is provided along with several unscripted natural living recordings, where the participant carried out a number of daily household activities which are annotated, where possible, throughout. Codes are provided to access the data and to replicate the ground-truthing procedure.

Cost-sensitive sequences of Bregman divergences.

The minimization of the empirical risk based on an arbitrary Bregman divergence is known to provide posterior class probability estimates in classification problems, but the accuracy of the estimate for a given value of the true posterior depends on the specific choice of the divergence. Ad hoc Bregman divergences can be designed to get a higher estimation accuracy for the posterior probability values that are most critical for a particular cost-sensitive classification scenario. Moreover, some sequences of Bregman loss functions can be constructed in such a way that their minimization guarantees, asymptotically, minimum number of errors in nonseparable cases, and maximum margin classifiers in separable problems. In this paper, we analyze general conditions on the Bregman generator to satisfy this property, and generalize the result for cost-sensitive classification.