The annual SAS Hackathon, which aims to generate solutions to real-world problems through the novel application of technology, has reached the adjudication state. The South African team of eight members from Dake Solutions has submitted a revolutionary diagnostic tool that could significantly improve medical outcomes for people with potentially life-threatening lung diseases.
Tasked with identifying a use case for advanced artificial intelligence (AI), machine learning and analytics that could create social value, Dake’s AI and Research team settled on the application of SAS environments and tools that would assist medical professionals to accelerate diagnoses of lung disease and prioritise high-risk individuals for treatment – especially in rural areas where lead times to diagnosis and treatment can be up to years long.
“Our specialty is advanced analytics, machine learning and AI. In this case, we relied on augmented intelligence to build models for disease identification. Though we focused on lung diseases for the Hackathon, these models have the potential to be scaled up to include other body parts for the same purpose of quick identification and prioritisation,” says Michael Austin, research developer at Dake Solutions.
The Dake team set out to create identification engines that could be trained through submitting positive and negative examples from X-ray images. “We focused on diseases that could be successfully identified using digital X-rays and we were fortunate that the models we constructed were successful – with limited data sets available the models were able to achieve diagnostic accuracy above 90%. Additional data sets to further train the models would raise that accuracy rate even further,” says Austin.
Jan Steenekamp, Enterprise Architect at Dake Solutions, says six seed models were created, associated with the most severe lung diseases. “From here, our intention is for further models to be created to cover other symptoms. Then, we can refine the reports the models produce so that someone with no medical knowledge can better understand the diagnostics from their X-ray image. It can also help to train medical students in lung disease diagnostics.”
This is at the heart of the social impact Dake’s solution seeks to deliver. “This will radically improve doctor-patient communication by laying the groundwork for three-dimensional models from digital X-rays using a coordinate-based system.”
The Dake team built a front-end web interface to allow users to upload their X-ray images. However, the team also considered ways to maximise the solution’s impact in rural areas where Internet connectivity is low and quick diagnosis would improve outcome statistics.
“These algorithms can ultimately be packaged into software that can run on a small computer device such as a Raspberry Pi. With a mobile X-ray facility, this means quick diagnostics and prioritisation of treatment can be achieved in the field, with more accurate readings obtained as and when Internet connectivity is available,” says Steenekamp.
Business analytics strategist Quintin De Klerk says this is just the starting point for the Dake team’s Hackathon concept. “These early diagnosis models can open up many more areas of research into diseases and deformations that will help radiologists and medical practitioners globally. Reports, heat maps and diagnostics can be produced literally within seconds, rather than the months to years it can currently take – especially in the diagnosis of lung cancer. That saved time in making informed decisions can mean the difference between life and death for patients with aggressive lung diseases, as well as improving the health and safety regimes of industries like mining, where diagnosis backlogs from regular lung screenings are a challenge.”
The Dake team “aimed big”, says De Klerk. Six weeks of intensive work has led to a number of models that are already highly accurate and can only improve in their diagnostic power.
“Using the SAS Viya Environment and Viya Model Manager allowed us to create intuitive machine learning architectures that underpinned the seed models. Once the models were uploaded, we were able to select from the challenger models to identify seed models and all were successful. We were then able to connect the model architecture to the front-end web page to enable uploads by users and diagnosis according to symptoms present in the X-rays,” De Klerk says.
“We are very thankful for the opportunity and enjoyed the journey, led by the expertise from the SAS team to cover our blind spots and help us get the most out of a challenging six weeks. This wouldn’t have been a success without their guidance using the tools. It was a humbling experience that has advanced our personal and professional growth,” says Austin.
Warren Murray, principal business solutions manager: business value at SAS EMEA, says: “This year was my first opportunity to be involved in the SAS Hackathon as a mentor – and it has been a memorable experience. This was largely due to the engagement with the team from Dake throughout the process. It has been inspiring to see what can be achieved in a very short space in time, by combining innovative ideas with SAS’s own capabilities and that of integration and orchestration of open source. The solution the Dake team has created has real world implications in bringing value to all players in the process.”
Refiloe Oliphant, channel and alliance specialist in South Africa, comments: “The hackathon is a strategic initiative that SAS uses to create a platform to help and support partners build solutions that differentiate themselves in a competitive market – a challenge that Dake has embraced and successfully completed within very tight timelines.”
From here, the Dake team’s submission will be judged alongside those from 69 other teams globally. The adjudication process will determine finalists by the end of May, based on how well the solutions address the use of technology, the processes followed, and the value brought to society through the application of the solution. The winning solutions for each category track will be announced in September 2022 at SAS Explore and will receive ongoing support from SAS to be brought to market in viable commercialised form.