Inferences from A&E data for resource optimisation in Brazilian hospitals
Over the last five years, Brazil has suffered a significant recession. Healthcare has been particularly affected with continuous budget cuts, and a large part of the general population is unable to get access to basic healthcare. The National Health Service in Brazil (SUS) has struggled to satisfy demand for essential health services, leading to unnecessary deaths in many cases. Over the last five years, 24,000 beds have been lost across public hospitals in Brazil.
The aim of this project was to analyse services and processes currently in practice at two hospitals in the state of Rio Grande do Sul, in order to identify areas of improvement. The work shown concerns Hospital Santa Cruz do Sul, a community hospital that is currently running at a deficit of BRL$1m per month. It serves a population of around 500,000 people in the region of Rio Pardo.
We identified two areas to focus on: (i) shift changes, and (ii) procedural changes such as room cleaning and calling patients to consultation.
An analysis of the given A&E data revealed that in both areas there is a potential for improvement. The impact of the proposed changes is modelled using SAN (Stochastic Automata Network) and Arena (queueing simulation). Although the simulation work is still ongoing, we currently estimate that implementing all proposed changes could lead to increasing the number of patients seen per doctor per hour by three during peak times. The benefits of this would be twofold: (i) the waiting time for patients would be reduced, thus potentially leading to better health outcomes, and (ii) less patients would walk away while waiting to be seen, which leads to more patients treated and more money recovered for the hospital while using the same level of staffing.
CanDL: Cancer Recognition with Deep Learning
Detection of cancer in medical images, e.g. from PET, MRI or CT imaging systems, is one of the core components in the fight against cancer. Currently tumour detection and segmentation is a very labour intensive job which requires the time of a highly trained person (costly), and is not 100% accurate, with a variety of factors (including human error and fatigue) diminishing the accuracy of the method. By automating this process clinicians will be greatly aided in their role, helping them to detect cancerous regions with a greater level of accuracy, and increasing the speed of detection to allow faster diagnosis, both leading to a much improved patient outcome, and helping the workload of these trained professionals, as well as decreasing the cost of the overall diagnosis process for the healthcare provider. The goal of this project is to develop an automatic tumour detection and segmentation of many tumour types in various parts of human body using deep learning.
A new phase in the development of cell tracking algorithms
In the field of cell biology, modern experimental and microscopy techniques generate vast amounts of high quality images of live cell populations over time. These datasets are so vast that they are indecipherable to a human observer in their raw form. This necessitates the development of automated computational algorithms that allow us to probe and understand as much of the available data as possible. Such algorithms must allow the tracking of potentially large cell populations, recovering dynamic quantities of interest such as speed and shape.
We describe two approaches to cell tracking developed in collaboration with Ibidi, a company that develop cell assays based in Munich, Germany. Specifically, we present a method for particle tracking applicable to large datasets that recovers the trajectories of cell centroids. We also describe a novel method for whole cell tracking, in which we recover entire cell morphologies using a phase field approach. A major advantage of the methodology, which appears new within the field, is that physical aspects of cell migration can be incorporated into the algorithm. Thus, hopefully generating more faithful approximation to the true morphologies.
The constraint modelling pipeline
Combinatorial search problems are ubiquitous across the public and private sectors, and academia. Consider a staff rostering problem to assign staff to shifts while respecting required shift patterns and staffing levels, physical and staff resources, and staff working preferences. The decision-making process is often further complicated by the need also to optimise an objective, such as to maximise profit or to minimise waste. We will deliver better solutions to these problems more rapidly, increasing efficiency and reducing cost.
It is natural to characterise such problems as a set of decision variables, each representing a choice that must be made in order to solve the problem at hand (e.g. which staff member is on duty for the Friday night shift), and a set of constraints describing allowed combinations of variable assignments (e.g. a staff member cannot be assigned to a day shift immediately following a night shift). A solution is an assignment of a value to each variable satisfying all constraints.
Many decision-making and optimisation formalisms take this general form. In all of these formalisms the model of the problem is crucial to the efficiency with which it can be solved. A model in this sense is the set of decision variables and constraints chosen to represent a given problem. There are typically many possible models and formulating an effective model is notoriously difficult. Therefore, automating modelling is a key challenge.
Over the last decade, in the context of Constraint Programming, we have taken a novel approach to addressing this challenge. The user writes a problem specification in the abstract constraint specification language ‘Essence’, capturing the structure of the problem above the level of abstraction at which modelling decisions are made. Our modelling pipeline, on which our proposed research is based, automatically generates a model from this specification. This removes the need for user constraint modelling expertise, and also preserves the structure of the specified problem, allowing the system easily to explore alternative models and to exploit properties such as symmetry.
Our pipeline generates constraint models equivalent in quality to those of a competent human constraint programmer, and so represents a significant milestone towards fully automated modelling.
Centre for Interdisciplinary Research in Computational Algebra (CIRCA)
With the increasing availability of smartphones, we sought to investigate whether:
- an app could be deployed with smokers intending to quit
- knowledge of craving and smoking behaviour helps smokers quit
Approximately 11,000 people die in Scotland each year from smoking related causes. While quitting smoking is relatively easy, maintaining a quit attempt is very difficult. Pharmaceutical treatments improve abstinence rates, however they do not address the spatial aspects of smoking behaviour.
Since smartphones can log spatial, as well as quantitative and qualitative data related to smoking behaviour, we can support smokers by first understanding their smoking behaviour and then sending dynamic support messages post-quit.
Dr Robert Schick
Dr Tom Kelsey
The robotic arm mimics the movement of the volunteer, detected via the muscle contraction sensor.
Wearable organic optoelectronic muscle contraction sensor
Wearable sensors are desirable for a wide range of medical applications including long-term continuous health monitoring and various robotic tools based on human-machine interactions. Ideally, wearable sensors should be light-weight, compact, flexible, non-invasive, easy to fabricate and low cost. Our work in organic optoelectronic devices suggests a new class of wearable sensors for medicine and sports. A flexible, non-invasive organic optoelectronic device is demonstrated which can be worn on the body to measure signals from muscles and control artificial limbs.
Organic optoelectronic devices exhibit attractive properties of tuneable light emission, easy fabrication on arbitrary, even flexible and stretchable, substrates, and the ability to generate and detect light. We have shown that the combination of organic light-emitting diodes (OLEDs) and organic photodiodes (OPDs) enables compact, non-invasive, flexible sensors for medical applications by simple solution-processing methods. Specifically, we developed a potentially low-cost muscle contraction sensor that can measure signals from intact muscles to control the movement of active prosthetic devices such as artificial limbs. Moreover, we demonstrated the feasibility of this all-organic, optoelectronic sensor by controlling a robotic arm so that it mimicked the motion of a healthy volunteer’s arm for two types of muscle contractions. The sensor measures the amount of light scattered by the skeletal muscle tissue along and parallel to the muscle fibre and can distinguish between isometric (i.e. muscle length remains constant) and isotonic (i.e. the muscle length changes, at a constant exerted force) contractions. Due to the ability to distinguish between different types of contractions, in addition to its non-invasive character and insensitivity to electromagnetic noise, the OLED and OPD based optical sensor is a promising alternative to the surface electromyography technique currently employed to detect the signals from muscles.
NOMAD (NMR Online Management and Datastore)
NMR (nuclear magnetic resonance) spectroscopy is most frequently used by chemists and biochemists to investigate the properties of organic molecules. The impact of this technique on the science has been substantial because of the range of information and the diversity of samples that can be analysed.
NOMAD (NMR Online Management and Datastore) is a cloud computing system developed through collaboration between the Schools of Chemistry and Computer Science at the University of St Andrews and has been funded through the EPSRC IAA funding. This system automates and simplifies a number of key workflows in NMR lab management, data acquisition and access which became recently a bottleneck for NMR data production.
Our recent development has been focused on interfacing NOMAD database with PURE (Research information system) in order to facilitate Open Access deposition of NMR data underpinning research at the School of Chemistry. Furthermore, the system architecture has been redone in order to make the transition from prototype (version 1.2 currently serving at the NMR facility at the University) to a market-ready product (version 2.0) that could be distributed to other NMR labs. We envisage that in future linking of NOMAD instances together could possibly create very useful Open Access NMR Data Repository.
Novel drug tools for neglected diseases: Development of tools for target identification of nitrofuran-carboxyamides with potent trypanocidal activity
Parasitic protozoan disease, such as African sleeping sickness, Chagas disease and leishmaniasis are some of the most neglected diseases in the world. The WHO have highlighted these diseases as a priority as a third of the world’s population are at risk; it effects millions of people and represents a huge percentage of the world’s disease burden. Current drug treatments are woefully inadequate, as many suffer from problems of toxicity, difficulty of administration in the field, cost and rapidly emerging parasite resistance. Therefore, there is an urgent need to identify novel therapeutic targets and to develop effective lead compounds that are safe, cheap and easy to administer against these protozoan diseases.
We have been optimising novel analogues of nifurtimox (a current front-line drug against some of these diseases), but with activity ~1000-fold more potent and without the side effects that nifurtimox has. Importantly, these novel analogues are relatively easy to synthesise and therefore cheap, and they are targeting a different, but still an essential biochemical process to those targeted by nifurtimox.
We are developing and utilising novel tools that allow us to elucidate the mode of action of our novel drugs, by identifying the protein target(s) they interact with and their subsequent metabolism. Collectively, these findings will have important implications for the future therapeutic treatment of African sleeping sickness, Chagas disease and possibly leishmaniasis. Our research involves utilising a multi-disciplinary approach spanning chemistry, biochemistry and molecular parasitology. The chemical technologies developed will be broadly applicable to protein target identification in general with the potential to simplify this challenging facet in the drug discovery process.
The figure at the top shows a stained trypanosome parasite and a chemical structure of one of the simplest chemical probes we are using to elucidate how our new compounds work within the parasite.
Single pixel compressive imaging
An image is worth a thousand words. This statement holds not only true in every day life but even more so in science. In a broader sense any position dependent measure can be understood as an imaging technique. One of the major challenges in imaging is correcting for aberrations, improving speed of imaging and reducing photo toxicity. The proof of concept that this project is developing aims to overcome, to some extent, these challenges. To do this we use structured illuminations and compressive sensing techniques. This approach aims at developing a toolbox (software and hardware) that can be used in medical and bio-photonics imaging context.
Optical Eigenmodes: http://www.eigenoptics.net/