Sydney 2017 Problems

PROBLEM

Sana Health Intelligence is an open Digital Health and Care platform to automate digital health and care.   The idea is to use affordable tech to automate care to reduce cost.    Sana starts from simple fitness tracker and android tablet to record wellness and interact with the users.

Sana will soon introduce Google Home speakers for voice interaction.     It is because a lot of seniors are less comfortable with screen but more comfortable with voice interaction.  At the moment, Care staff must visit the seniors with a clip board to conduct manual subjective assessment and interaction.    Any automation could bring huge saving to our aged care system.

SOLUTION

Using a Google Home Mini to create an engaging voice interaction with a seniors.   The idea could be: 

1.  Engage them in simple conversation or game to overcome isolation, turn on music.

2.  Engage and motivate them to take action, walk,  medication reminder etc etc.

3.  Send text message or voice message to care staff or family.  Receive/ Listen message from care staff or family 

4.  Initiative for interactive voice assessment of well being,  health condition or care services.

PROBLEM

We have recently started to use a wrist-based wearable device and smartphone to obtain second-by-second recoding of heart rate and global positioning system (GPS) coordinates. By combining these two sources of data, we can monitor changes in heart rate before, during and after various physical activities. This has important potential application in terms of early detection of heart failure. Traditionally, these data are visualised as separated tracks on paper or on screen. It is difficult for clinicians to impressively experience the patient’s dynamic changes of heart rate. 

SOLUTION

To help clinicians visualise the relationship between the physical activity (running/walking/stopping/uphill/downhill….) and heart rate changes, we propose to develop a 3D virtual reality visualization platform as a mobile app or a mobile-enabled web browser, which could be used in conjunction with a low-cost VR headset (e.g., Google Cardboard). We believe this tool will help clinicians directly experience the patient’s heart rate changes in response to different physical activity.

PROBLEM

 The Victor Chang Cardiac Research Institute has been using Optical Projection Tomography (OPT) for obtaining 3D images of embryonic hearts from various genetic mouse models of congenital heart disease. Three dimensional imaging techniques like OPT is a powerful means of identifying fine anatomical abnormalities in the heart.  Nonetheless, the reconstructed 3D images can often be quite noisy, and therefore it can be difficult to identify fine structural abnormalities.

SOLUTION

Recent advances in image denoising techniques have shown success in refining the quality of images. We believe being able to run one or more of state-of-the-art demonising algorithms (e.g., https://github.com/albarji/proxTV) to process our 3D images can improve the overall quality of the images and aid discovery of structural defects from these images.

PROBLEM

I (Brenda Castro Pelayo) work as Design Lead for product development at Hello Sunday Morning, a non-for-profit that helps people change their relationship with alcohol and live healthier lives. 

One of the biggest problems we see around our field of work, is how difficult is to successfully help a loved one start the process of change. Simply starting a conversation about alcohol consumption with someone with a problematic drinking behaviour can be very difficult and stressing, most of the times leading to situations contrary as planned.

People wanting to help their loved ones change their drinking habits usually don't know how to motivate and give support in the right way for a specific person in a specific state of change. Instead, they can end up creating distance and mistrust in the relationship. Helping someone make the choice to change drinking habits is not easy, it requires empathy, patience, understanding, and touch. It requires access to the right sources of information about alcohol consumption and addiction, access to advice from experts and sometimes from people going through similar experiences.

SOLUTION

At Hello Sunday Morning we have developed a program, accessible through mobile phone, that supports people through their process of changing their drinking habits. This program, called Daybreak, has been available since 2016 and has helped 2480 people change their relationship with alcohol and has successfully improved the mood of 9400 people, which is an important factor of success in changing a habit. The program works based on three factors: A community of support with people going through similar experiences, personal health coaching by professional therapists, and a collection of strategies that we call experiments to help them through their change process. 

This program focuses on helping people who are already in a stage of being ready to change. About 49% of Australians say they want to change their relationship with alcohol, but not all who say they want to change are actually ready to do so. Many of them might need some peer support from their family and friends to choose to join a program like Daybreak. And there is the other 51%, from which of course it is only a few for whom drinking has become problematic, but it is an important number in any case. 

We would like you to design and prototype a solution that helps family and friends of people with a problematic drinking to be better at supporting their loved ones and help them become ready to change.

Check out our information page about this problem yet to be solved: https://www.hellosundaymorning.org/daybreak/family-and-friends/

PREVIOUS SOLUTIONS

Psychological therapy (e.g. couples), some treatment programs like AA that may extend to family and friends, addiction treatment and rehab centres. 

PROBLEM

There is significant support for both the research community as well as industry focused in developing new products or services in the Life Sciences (medical devices, pharmaceuticals, biotechnology) sector in Australia; however, a single repository of this information does not exist, and keeping the information current/up-to-date is difficult. There are over a 100 different granting schemes, both Federal and State-based that support the life sciences (ARC, NH&MRC, Department of Industry Entrepreneurs Program/Accelerating Commercialisation, Global Innovations Linkages, ATSE’s Global Connect, as well as may State-based programs/schemes). For companies that haven’t previously invested in R&D, or been aware of the support available, searching through the various databases can be daunting and very confusing- the search tools are not user-friendly.

SOLUTION

We envision the development of a “grants portal” which allows users to easily search through all the available granting schemes in Australia. The portal would ideally ask the user a series of profile questions (drop-down menus), which will help filter the information based on the grants’ eligibility criteria (ie size of company, revenue, matching funds requirements, etc). This way, only the granting schemes that are relevant to the user will be presented. Such tool would be a very significant development and benefit for the life sciences sector in Australia, providing an avenue for growth and increased employment. The portal would ideally provide a functionality to assist the user in completing the grant application, with prompts for missing info, etc…

As stated on the problem definition, links/databases for existing schemes may be out of date, so, ensuring that the user is prompted with the most up-to-date information is a must (the portal to query the source info and display most recent release/version date). The portal should pull info from existing databases and be “notified” of changes to links to the information.

PROBLEM

Genome.One is developing an automated clinical pipeline to efficiently sequence the DNA of clinical patients, but at present there remains a single manual step which still needs to be done, and grinds the process to a halt, forcing us to wait on a human operator to perform a single human-intelligence task. This human-intelligence task is to note down the relationships between family members in a clinical case, something we typically have to guess at given only the patient names and dates of birth. Because of the time-sensitive nature of the clinical cases which are waiting on our results, many of whom are in life-threatening danger even before the test is ordered, the delay introduced by this manual check can be very costly.

SOLUTION

We believe that the job of determining relationships between persons given only their names and dates of birth can be done by a computer just as easily as by a human, and far more rapidly. Further, this would likely reduce human error, by applying consistent rules and standardising the method used. Solving this problem would remove the last manual bottleneck from our process, leading to full automation of the data processing pipeline used in our clinical genomics cases.

There are, however, a number of edge cases which need to be considered (e.g. a group of patients who are brothers-in-law but with different last names), and the relationships derived need to also be coded into clinical standard metadata (to allow data to be de-identified, and have hospital systems correctly pick up the nuances of the relationships in question). The problem owner can provide guidance as to the kinds of situations typically encountered, and how they are approached by human operators in at present.

PREVIOUS SOLUTIONS

At present, we are not aware of a solution that automates this step. Our solution, in practice, is to solve the problem via manual intervention.

PROBLEM

The problem that we want to apply AI to is physiotherapy. A physiotherapist has to track the progress of range of motion (ROM) abilities in patients, e.g. during rehab post a surgery, which is typically done by measuring angles of limbs. Often times, physios note the progress by measuring angles and do so by estimating the angle, which may result in subjective estimates of the progress of a patient rather than objective measured results.

SOLUTION

The idea is to use a video camera to capture a person’s movement and objectively calculate the angles between their limbs over the course of several therapy sessions and to automatically write a report from this. This would help a physio become more objective in their work and do less paperwork at the same time. Ideally, the ROM analysis would be done in real-time during a live consultation, which could be held online.

The demonstrator would be able to use existing human pose estimation algorithms that are quite reliable. Some open source approaches exist:, e.g. https://github.com/eldar/pose-tensorflow or http://www.robots.ox.ac.uk/~vgg/data/pose/ which can be used in this project to put together a demonstrator for real-time range-of-motion analysis of limbs. Making it work in a Web browser would be a bonus. We can then plug it into Coviu for actual use by clinicians.

PREVIOUS SOLUTIONS

Physios typically estimate limb angles by inspection. Sometimes they take a photo and measure it on the computer, but rarely are the notes documented with short videos.

PROBLEM

The field of precision medicine uses genome sequencing to identify the genetic mutations in each patient's cancer, which can be matched to targeted drugs which are highly effective against specific mutations. However, mostly these drugs are novel and are are only available to patients through clinical trials, and the landscape of trials, gene targets, and drugs is constantly evolving. We need to develop a solution to help match patients with relevant trials.

SOLUTION

Clinical trials are registered at ANZCTR (the Australian clinical trials registry) in a semi-structured, free text format. The scope of the hack is to extract the information about each cancer clinical trial (e.g. XML file download), and translate important information (e.g. drug, gene, cancer type) into a machine readable form using text mining, ideally in a way which can be constantly kept up to date. Furthermore we wish to build an API and a website to access this machine readable information. Individual clinical trial records may have incomplete information, so additional resources like which gene a given drug targets may need to be obtained from drugbank.ca.

CURRENT SOLUTIONS

IBM Watson offers a commercial (and costly) ClinicalTrials matching service, and molecularmatch.com offers a similar service in the USA. However, there is currently no effective solution for Australian clinical trials. We currently download the XML files from ANZCTR, and search the free text, on demand, which hampers our ability to automate the treatment recommendations for the dozens of patients we analyse each week.

PROBLEM

Infertility can be hard to diagnose because in many cases there is more than one potential reason detected for the couple or person seeking treatment, and in as many as 20-30% of cases there is no apparent reason. There are several different possible treatments for infertility that can be selected based on the type of infertility, and their personal preferences around the type of procedures they feel comfortable undergoing. In particular infertility that might result from a previous sexually transmitted infection is very hard to communicate in a positive manner.

SOLUTION

Using a sample set of fertility and laboratory data based on a human cohort that is both qualitative and quantitative we wish to build some visualisations and scientific communication around infectious ideologies and infertility. Using blood markers and other data we hope to help patients make informed decisions about their treatment, whether to save money for IVF for instance, and understand how their conditions work through data storytelling. 

The primary outcomes from the solution will ideally be

1. Visually depict data features that represent a match between the biological and the fertility data to try and simplify the complexity surrounding infertility issues

2. Communicate scientific test results and participant data sets together in an easily digestible format for patients and concerned parties.

PREVIOUS SOLUTIONS

Infertility diagnosis is often discussed in broad ways for the different kinds of infertility. Infertility that might have resulted from sexually transmitted infections is often referred to as tubal factor infertility because of the damage to the tubes and to reduce the stigma. Diagnosis is typically by an invasive costly surgery (laparoscopy) and the most successful fertility treatment is IVF. Surgery is not often performed, so women with this form of infertility often undergo other less successful treatment.

PROBLEM

Building bioinformatics solution that satisfy both demands can be challenging. While our genome engineering web service, GT-Scan2, satisfies the speed requirement, the Health Hack community can contribute towards increasing the reproducibility of the framework.

We think of GT-Scan2 as the search engine for the genome, where researchers can type in the gene they want to edit and GT-Scan2 returns a list of optimal interaction points in the DNA that are close to the gene of interest. It is one of the first complex serverless architectures using AWS’s new Lambda functions and as such received international attention (e.g.https://www.genomeweb.com/informatics/australian-team-puts-crispr-design-amazon-cloud).

SOLUTION

As part of the Health Hack, we want to extend its functionality to allow researchers to document their process of choosing the optimal target site and hence increase the reproducibility of their work. More specifically, researchers should be able to trigger GT-Scan2 runs from within their documentation framework, be that a slack channel or a Jupyter notebook through GT-Scan2’s API. Another API call will then retrieve the results after the run has finished.

The challenge: Unlike other APIs, which return results back to notebooks (Cognoma API) , GT-Scan2 consists of massively parallel serverless functions which return results to the webpage independently, as a result the API ‘GET’ function does not know when this process has finalized. We therefore need to implement an AWS messaging service that can inform the GET function when to retrieve the data. Furthermore, for Slack support a function needs to be designed through “Slash commands” plugin to communicate GT-Scan2 API and return the response information as human-readable text to Slack.

PROBLEM

The problem is that medical terminology is not really standardised, ‘large head’, ‘big head’ and ‘macrocephaly’ all mean the same thing, but this would not be obvious to a computer. You could not even tally up all people with ‘large head’ in a spreadsheet if some people have been recorded as having a ‘big head’. Let alone do anything more fancy, like machine learning.

The Human Phenotype Ontology (HPO) has been developed to address this problem. The HPO provides a standard vocabulary of over 10’000 terms to describe clinical features of patients with rare diseases. For example HPO Term [HP:0000256] stands for ‘macrocephaly’ and has the synonyms ‘large head’ and ‘big head’.  If everybody used HPO terms to describe clinical features of their patients, research on rare genetic diseases would progress in leaps and bounds!

Unfortunately, looking up the right HPO term for a clinical feature takes a few extra steps: you need to go to the HPO website, search for your clinical feature of interest (e.g. ‘large head’),  choose the right HPO term from the search results, copy the HPO ID and paste it into your primary database. Though not particularly onerous, having to do this hundreds of times quickly becomes time consuming. I would like to build a browser plugin that facilitates this process. Just highlight a clinical feature in your browser, click the HPO extension button and your clipboard will contain the correct HPO ID ready to paste wherever you need it!

SOLUTION

Bioportal and the EBI Ontology Lookup Service provide APIs to query the HPO, which could be used for this project. In fact, these APIs can query many other biomedical ontologies, so the Chrome Extension could be easily modified to serve other researcher needs! 

Onomaton is a current solution but it only works in Google Sheets and has some other limitations.