Search

Background and aims: Using reports of randomised trials of smoking cessation interventions as a test case, this study aimed to develop and evaluate machine learning (ML) algorithms for extracting information from study reports and predicting outcomes as part of the Human Behaviour Change Project. Methods: Researchers manually annotated 70 items of information (‘entities’) in 512 reports of randomised trials of smoking cessation interventions covering intervention content and delivery, population, setting, outcome and study methodology using the Behaviour Change Intervention Ontology. These entities were used to train ML algorithms to extract the information automatically. The information extraction ML algorithm involved a named-entity recognition system using the ‘FLAIR’ framework. The manually annotated intervention, population, setting and study entities were used to develop a deep-learning algorithm using multiple layers of long-short-term-memory (LSTM) components to predict smoking cessation outcomes. Results: The F1 evaluation score, derived from the false positive and false negative rates (range 0-1), for the information extraction algorithm averaged 0.42 across different types of entity (SD=0.22, range 0.05 to 0.88) compared with an average human annotator’s score of 0.75 (SD=0.15, range 0.38 to 1.00). The algorithm for assigning entities to study arms (e.g., intervention or control) was not successful. The ML outcome prediction algorithm did not outperform prediction based just on the mean outcome value or a linear regression model. Conclusions: While some success was achieved in using ML to extract information from reports of randomised trials of smoking cessation interventions, we identified major challenges that could be addressed by greater standardisation in the way that studies are reported. Outcome prediction from smoking cessation studies may benefit from development of novel algorithms, e.g., using ontological information to inform ML.

In this article, we provide an evidence-based primer of current tools and evolving concepts in the area of intraprocedural artificial intelligence (AI) methods in colonoscopy and laparoscopy as a ‘procedure companion’, with specific focus on colorectal cancer recognition and characterisation. These interventions are both likely beneficiaries from an impending rapid phase in technical and technological evolution. The domains where AI is most likely to impact are explored as well as the methodological pitfalls pertaining to AI methods. Such issues include the need for large volumes of data to train AI systems, questions surrounding false positive rates, explainability and interpretability as well as recent concerns surrounding instabilities in current deep learning (DL) models. The area of biophysics-inspired models, a potential remedy to some of these pitfalls, is explored as it could allow our understanding of the fundamental physiological differences between tissue types to be exploited in real time with the help of computer-assisted interpretation. Right now, such models can include data collected from dynamic fluorescence imaging in surgery to characterise lesions by their biology reducing the number of cases needed to build a reliable and interpretable classification system. Furthermore, instead of focussing on image-by-image analysis, such systems could analyse in a continuous fashion, more akin to how we view procedures in real life and make decisions in a manner more comparable to human decision-making. Synergistical approaches can ensure AI methods usefully embed within practice thus safeguarding against collapse of this exciting field of investigation as another ‘boom and bust’ cycle of AI endeavour.

Background Artificial Intelligence (AI) is transforming the process of scientific research. AI, coupled with availability of large datasets and increasing computational power, is accelerating progress in areas such as genetics, climate change and astronomy [NeurIPS 2019 Workshop Tackling Climate Change with Machine Learning, Vancouver, Canada; Hausen R, Robertson BE. Morpheus: A deep learning framework for the pixel-level analysis of astronomical image data. Astrophys J Suppl Ser. 2020;248:20; Dias R, Torkamani A. AI in clinical and genomic diagnostics. Genome Med. 2019;11:70.]. The application of AI in behavioral science is still in its infancy and realizing the promise of AI requires adapting current practices. Purposes By using AI to synthesize and interpret behavior change intervention evaluation report findings at a scale beyond human capability, the HBCP seeks to improve the efficiency and effectiveness of research activities. We explore challenges facing AI adoption in behavioral science through the lens of lessons learned during the Human Behaviour-Change Project (HBCP). Methods The project used an iterative cycle of development and testing of AI algorithms. Using a corpus of published research reports of randomized controlled trials of behavioral interventions, behavioral science experts annotated occurrences of interventions and outcomes. AI algorithms were trained to recognize natural language patterns associated with interventions and outcomes from the expert human annotations. Once trained, the AI algorithms were used to predict outcomes for interventions that were checked by behavioral scientists. Results Intervention reports contain many items of information needing to be extracted and these are expressed in hugely variable and idiosyncratic language used in research reports to convey information makes developing algorithms to extract all the information with near perfect accuracy impractical. However, statistical matching algorithms combined with advanced machine learning approaches created reasonably accurate outcome predictions from incomplete data. Conclusions AI holds promise for achieving the goal of predicting outcomes of behavior change interventions, based on information that is automatically extracted from intervention evaluation reports. This information can be used to train knowledge systems using machine learning and reasoning algorithms., Artificial intelligence holds promise for achieving the goal of predicting outcomes of behavior change interventions, using information automatically extracted from intervention evaluation reports.

Findings from randomized controlled trials (RCTs) of behaviour change interventions encode much of our knowledge on intervention efficacy under defined conditions. Predicting outcomes of novel interventions in novel conditions can be challenging, as can predicting differences in outcomes between different interventions or different conditions. To predict outcomes from RCTs, we propose a generic framework of combining the information from two sources - i) the instances (comprised of surrounding text and their numeric values) of relevant attributes, namely the intervention, setting and population characteristics of a study, and ii) abstract representation of the categories of these attributes themselves. We demonstrate that this way of encoding both the information about an attribute and its value when used as an embedding layer within a standard deep sequence modeling setup improves the outcome prediction effectiveness.

The wide availability of near infrared light sources in interventional medical imaging stacks enables non-invasive quantification of perfusion by using fluorescent dyes, typically Indocyanine Green (ICG). Due to their often leaky and chaotic vasculatures, intravenously administered ICG perfuses through cancerous tissues differently. We investigate here how a few characteristic values derived from the time series of fluorescence can be used in simple machine learning algorithms to distinguish benign lesions from cancers. These features capture the initial uptake of ICG in the colon, its peak fluorescence, and its early wash-out. By using simple, explainable algorithms we demonstrate, in clinical cases, that sensitivity (specificity) rates of over 95% (95%) for cancer classification can be achieved.

Due to the fast pace at which randomized controlled trials are published in the health domain, researchers, consultants and policymakers would benefit from more automatic ways to process them by both extracting relevant information and automating the meta-analysis processes. In this paper, we present a novel methodology based on natural language processing and reasoning models to 1) extract relevant information from RCTs and 2) predict potential outcome values on novel scenarios, given the extracted knowledge, in the domain of behavior change for smoking cessation.

In order of presentation at the conference

Dual emissions at ~700 and 800 nm have been achieved from a single NIR-AZA fluorophore 1 by establishing parameters in which it can exist in either its isolated molecular or aggregated states. Dual near infrared (NIR) fluorescence color lymph node (LN) mapping with 1 was achieved in a large-animal porcine model, with injection site, channels and nodes all detectable at both 700 and 800 nm using a preclinical open camera system. The fluorophore was also compatible with imaging using two clinical instruments for fluorescence guided surgery. Methods: An NIR-AZA fluorophore with hydrophilic and phobic features was synthesised in a straightforward manner and its aggregation properties characterised spectroscopically and by TEM imaging. Toxicity was assessed in a rodent model and dual color fluorescence imaging evaluated by lymph node mapping in a large animal porcine models and in ex-vivo human tissue specimen. Results: Dual color fluorescence imaging has been achieved in the highly complex biomedical scenario of lymph node mapping. Emissions at 700 and 800 nm can be achieved from a single fluorophore by establishing molecular and aggregate forms. Fluorophore was compatible with clinical systems for fluorescence guided surgery and no toxicity was observed in high dosage testing. Conclusion: A new, biomedical compatible form of NIR-dual emission wavelength imaging has been established using a readily accessible fluorophore with significant scope for clinical translation.

Time analysis of the course of an infectious disease epidemic is a critical way to understand the dynamics of pathogen transmission and the effect of population scale interventions. Computational methods have been applied to the progression of the COVID-19 outbreak in five different countries (Ireland, Germany, UK, South Korea and Iceland) using their reported daily infection data. A Gaussian convolution smoothing function constructed a continuous epidemic line profile that was segmented into longitudinal time series of mathematically fitted individual logistic curves. The time series of fitted curves allowed comparison of disease progression with differences in decreasing daily infection numbers following the epidemic peak being of specific interest. A positive relationship between the rate of declining infections and countries with comprehensive COVID-19 testing regimes existed. Insight into different rates of decline infection numbers following the wave peak was also possible which could be a useful tool to guide the reopening of societies. In contrast, extended epidemic timeframes were recorded for those least prepared for large-scale testing and contact tracing. As many countries continue to struggle to implement population wide testing it is prudent to explore additional measures that could be employed. Comparative analysis of healthcare worker (HCW) infection data from Ireland shows it closely related to that of the entire population with respect to trends of daily infection numbers and growth rates over a 57-day period. With 31.6% of all test-confirmed infections in healthcare workers (all employees of healthcare facilities), they represent a concentrated 3% subset of the national population which if exhaustively tested (regardless of symptom status) could provide valuable information on disease progression in the entire population (or set). Mathematically, national population and HCWs can be viewed as a set and subset with significant influences on each other, with solidarity between both an essential ingredient for ending this crisis.

Background Surgery is a major component of health-care provision. Operative intervention often employs minimally invasive approaches incorporating digital cameras creating a ‘digital twin’ of both intracorporeal appearances and operative performance. Video recordings provide richer detail than the traditional operative note and can couple with advanced computer technology to unlock new analytic capabilities capable of driving surgical advancement via quality improvement initiatives and new technology design. Surgical video is however an under-utilized technology resource, in part, because ownership along with broader issues including purpose, privacy, confidentiality, copyright and inclusion in outputs have been poorly considered using outdated categorisation. Method A first principles perspective on operative video classification as a useful public interest resource enshrining fundamental stakeholder (patients, physicians, institutions, industry and society) rights, roles and responsibilities. Result A facility of noble purpose, understandable to all, for fair, accountable, safe and transparent access to large volumes of anonymised surgical videos of intracorporeal operations that enables advances through cross-disciplinary research is proposed. Technology can be exploited to protect all relevant parties respecting both citizen data-rights and the special status doctor-patient relationship. Through general consensus, the capability can be understood, established and iterated to perfection. Conclusion Overall we argue that new and specific classification of surgical video enables responsible curation and serves the public good better than the current model. Rather than being thought of as a bicycle where discrete ownership is ascribed, such data are better viewed as being more like a park, a regulated amenity we should preserve for better human life.

Changing behaviour is necessary to address many of the threats facing human populations. However, identifying behaviour change interventions likely to be effective in particular contexts as a basis for improving them presents a major challenge. The Human Behaviour-Change Project harnesses the power of artificial intelligence and behavioural science to organise global evidence about behaviour change to predict outcomes in common and unknown behaviour change scenarios.

Our understanding of human behavior and how to effect changes in behavior have evolved to the point where behavior change techniques have been applied successfully in areas as diverse as public health, advertising and retail. Despite its successes, behavior change science remains largely unknown among the engineering and technology community. In this chapter we provide a brief introduction to a vast field of knowledge. We describe practical behavior change tools and techniques that can be used to design behavioral interventions as elements of complex projects such as are typical of the sharing economy. Structured design of behavioral change programs is discussed and key concepts discussed followed by an overview of important techniques such as the COM-B Model, the Behavior Change Wheel. The Taxonomy of Behavior Change Techniques we discuss sits in an exciting area of active research using ontology techniques to organize knowledge. This approach lends itself naturally to application of technological tools such as artificial intelligence. Behavior change is a rapidly evolving science with sufficient rigorous and structured underpinnings to support more holistic design of complex projects where human behaviors are critical to sustainability and long-term success. We invite the reader to see this chapter as a practical introduction and to use the accompanying references to explore the literature around this fascinating subject.

The Laplace mechanism is the workhorse of differential privacy, applied to many instances where numerical data is processed. However, the Laplace mechanism can return semantically impossible values, such as negative counts, due to its infinite support. There are two popular solutions to this: (i) bounding/capping the output values and (ii) bounding the mechanism support. In this paper, we show that bounding the mechanism support, while using the parameters of the standard Laplace mechanism, does not typically preserve differential privacy. We also present a robust method to compute the optimal mechanism parameters to achieve differential privacy in such a setting.

Colorectal anastomotic leakage remains a serious complication with implications on hospital stay, oncological outcomes(1) and treatment cost. (2) Traditional intra-operative visual perfusion assessment has been shown to be suboptimal(3) and surgeons are looking to indocyanine green (ICG) fluorescence angiography as an adjunct to clinical judgement.

This paper describes our approach to construct a scalable system for unsupervised information extraction from the behaviour change intervention literature. Due to the many different types of attribute to be extracted, we adopt a passage retrieval based framework that provides the most likely value for an attribute. Our proposed method is capable of addressing variable length passage sizes and different validation criteria for the extracted values corresponding to each attribute to be found. We evaluate our approach by constructing a manually annotated ground-truth from a set of 50 research papers with reported studies on smoking cessation.

Person-specific data offer enormous opportunities for deriving insights that can radically improve different facets of our everyday lives, ranging from the provisioning of personalized medicine and healthcare, to the offering of smart transportation and smart energy. At the same time, the use of person-specific data to support these applications can come at a high cost to individuals' privacy, unless proper de-identification technology is in place to provide rigorous privacy guarantees. In this paper we introduce PRIMA, an end-to-end solution allowing decision makers to map out and execute their data privacy strategy through a comprehensive workflow. Our toolkit offers an intuitive risk-utility exploration framework for end users to navigate through the enormous number of possible combinations of anonymization settings and provide meaningful reports that help them understand the impact of each strategy in terms of utility and risk. Unlike traditional approaches, that rely on limited scale tools and manual analyses, our toolkit is the first scalable, production-grade system that can execute all of its components (such as vulnerability analysis, anonymization, risk and information loss measurements) on arbitrarily large datasets. Furthermore, it offers a flexible library for developers to integrate and extend its functionality to embed de-identification components into their applications.