Dr CADx - AI Medical Imaging Diagnostics

Problem: Because radiologists are human, they are not 100% perfect in interpreting medical, and studies have quantified their diagnostic accuracy to be 70% on average. The situation is even more dire in developing regions like Africa where there is a serious scarcity of radiologists. Patients are thus frequently misdiagnosed resulting in them failing to get the right treatment on time, leading to undue suffering and even death.

Solution: Dr CADx, through a computer aided diagnostic system (CAD) that helps doctors diagnose medical images more accurately and at a low cost, will help reduce the human errors that have dire consequences for patients and drive up medical costs. 

There is a serious scarcity of radiologists, especially in developing countries. E.g. in Zimbabwe there are only 17 radiologists serving a population of about 16 million giving a ratio of just one radiologist for every million people, as compared to 1: 9,000 for the US. 16 of these radiologists are based in the capital Harare and one is in the second city Bulawayo. Additionally, none of these are employed in the public hospitals that serve most of the population, and this scenario is the same across many African countries, and some countries don’t even have a single radiologist. As a result, annually an estimated 45 million patients in Africa do not have access to a radiologist to review their medical images.

Additionally, even in the developed world, because radiologists are human, they are not 100% perfect, and studies have quantified their diagnostic accuracy to be 70% on average. As a result of the 3.6 billion medical images that are taken each year, about 1 billion scans are misdiagnosed resulting in patients frequently failing to get the right treatment on time, leading to undue suffering and even death.

Dr CADx is using state-of-the-art deep learning algorithms to develop an application that can rapidly and accurately interpret medical images at a low cost. Deep learning is a technology that simulates how the human brain works to teach itself by learning from a large set of exemplars, in this case to recognise disease patterns in images. With an easy to use application the doctor simply uploads the image to Dr CADx servers and in less than a minute gets the diagnosis results with a good accuracy

Our target population is two fold, 1. every person who gets a medical image taken and 2. the doctors who interpret the images especially the non radiology doctors. Doctors have informed our decision making process from the begining through one on one conversations, focus groups and having one as part of our company helping tailor the solution, we strive to understand their needs and how best to package then in the most effective way. The imaged population needs the best possible results at the most affordable cost in a timely manner and they communicate this strongly.

Dr CADx will address the doctors' and imaged populations needs by providing a highly accurate diagnostic service that consistantly gives high quality results at all hours of the day, including in remote areas at affordable prices meaning that both doctors and their patients recieve immediate quality diagnostic results that enable the best healthcare to be provided resulting in faster routes to wellness.

Dr CADx improves access to affordable healthcare, providing quick diagnostic services where they are none, enabling faster disease treatment, reducing severity of disease progression. This reduces treatment costs, disease progression complexities due to wrong, delayed treatment and spread in communities.

Inherent in Dr CADx is data for disease surveillance, virtually absent in Africa, leveragable to inform decision making to improve heathcare supply chains. Images uploaded onto Dr CADx cloud with different diseases will be geographically mapped, used to track spread of emerging outbreaks, analyzed to show trends, timelines and numbers in disease spread and used for rapid response to pandemics.

We are making use of AI to develop a software solution that augments the capabilities of doctors to achieve more accurate interpretation of medical images. By developing our own unique ensemble of image processing and deep learning algorithms that achieve high accuracy of 96% for TB, 85% on COVID-19 and 84% on 14 chest findings that include pneumonia consolidation, lung cancer nodules/masses and pneumothorax as the most common ones.

In addition to making use of data sourced from the US, Europe and Asia like our key competitors Enlitic (US), Delft Imaging (Netherlands) and Lunit (South Korea), we are also acquiring significant data within Africa. This is to ensure good generalisation of our algorithms across all demographics, and more specifically it will guarantee validated performance within our initial target African market. 

We are putting strong focus on making it accessible offline as well as online. This will ensure Dr CADx caters for the many hospitals and clinics with limited internet connectivity across Africa. Given the much more severe shortage of radiologists in Africa, we have designed the tool to meet the needs of general practitioners to diagnose a range of conditions.

So by deploying an AI system that will be available at any given time of the day — even in remote areas — we will greatly improve the quality of patient management, reduce costs and save lives, especially in the marginalized areas of the developing world. Dr CADx will thus revolutionize the healthcare system, benefiting patients, doctors and hospitals alike.

We are making use of image recognition technology to develop our computer aided diagnosis software platform. We are developing machine learning algorithms based on deep convolution neural networks that can learn the patterns in medical images that are characteristic of the various pathologies.

Deep learning is a technology that simulates how the human brain works in identifying patterns in images. Like a human doctor who learns from seeing several example cases, the process of training Dr CADx involves supplying it with a large set of medical images that have already been annotated by experts. The system analyses the images and automatically learns the features in the images that are characteristic of each disease.

The algorithms are deployed as a either a) a cloud based web app, b) an API to integrate with other health care systems, c) a desktop or mobile app that also work in offline environments and d) software embedded into medical imaging equipment. 

A pilot validation study was conducted to ascertain the standalone accuracy of Dr CADx, the accuracy of doctors when they interpret chest X-rays for TB in the traditional way and their accuracy when they are aided by Dr CADx.

The standalone accuracy of Dr CADx against the GeneXpert gold standard was 96%. 

The mean accuracy of the 9 unaided doctors was 88%, and it increased to 90% when they used Dr CADx showing a 2% accuracy improvement.

For development stage validation for COVID-19 we did a standalone test on 93 chest X-rays sourced from a completely different hospitals that the images used for training and achieved an 85% accuracy. 

A standalone test for the other 14 chest findings yielded an average accuracy of 84% which is comparable to expert radiologist level performance.

A video of the product demo can be viewed here.

Our deployment of Dr CADx for in the diagnostic workflow will result in outputs of doctors achieving improved accuracy in interpreting medical results. In a pilot study we conducted that involved 9 doctors using Dr CADx, the doctors were seen to be 4% more accurate when using Dr CADx compared to when they didn't use the software. 

The ensuing short term outcomes are that this will result in correct treatment being given to more patients right from the start lead to better patient outcomes as complications due to increased complexity of disease are reduced. Additionally this will see a reduction overall cost of treatment. A study by Walsh-Kelly CM et. al. quantified the economic costs of misinterpretation of X-rays to be $85.

The long term outcomes will be that patients will thus live longer better lives and a reduction in the mortality rates. In the case of communicable diseases like tuberculosis and COVID-19 the speed and higher accuracy of diagnosis will result in reduced spread of the disease leading to reduced socioeconomic disease burden.

Currently our solution is not yet in clinical use and hence has not yet started serving patients. We are planning to start deploying it in the last quarter of this year and expected to have served 17,000 patients within a year. In five years time, in 2025, we forecast our solution will analyse medical images for 7.3 million patients in that year. Cumulatively from now until 2025 Dr CADx will have served 11 million patients. 

Goals for next 12 months

1) Complete development of the full stack, i.e. algorithms, back-end APIs and front-end user interfaces to achieve the minimum viable product (MVP).

2) Clinically validate the system with doctors for 16 chest findings.

3) Market introduction of our solution by the last quarter of 2020 leading to us serving 17,000 patients mainly from local income populations in Sub-Saharan Africa.

4) Establish distribution agreements with at least one organisation each in Zambia, DRC and Nigeria to accelerate deployments in those countries based on their existing experience and relationships with customers and users there.

5) Execute a collaboration agreement with at least one medical equipment manufacturer to integrate our solution into their machines.

Goals for next 5 years

Our ambition is that in 5 years time our solution would have been used to analyse medical images for 11 million patients with greater accuracy that is possible today. By enabling doctors to make more accurate diagnosis earlier, it will lead to more positive treatment outcomes for these millions of patients. We plan to achieve this by:

1) Collaborating with Ministries of Health, heathcare IT solution providers and other international organisations to replicate and scale up in the local markets. 

2) Expand the imaging modalities we work with to include CT scans (2023), mammography and ultrasound (2024) and MRI (2025).

3) Expand the countries we operate in to cover at least half of Africa by 2023 and expand internationally into Europe and South America by 2024.

1) High capital requirements to complete product development and undertake clinical validation.

2) Different and sometimes unclear regulatory requirements in the various African markets we want to enter.

3) Long-time to get FDA/CE approval which is essential for wider market acceptance and scaling up.

4) Low purchasing power in poor communities which are among the ones that can benefit the most from our solution due to the much lower availability of radiology expertise there.

1) Continuous fundraising activities for both non-dilutive grants and private equity investment.

2) Start by entering markets with less rigorous regulatory requirements and progressively work on procedures for compliance in the others.

3) Initially target early adopter customers whilst pursing FDA/CE approval to satisfy other customers.

4) Collaborate with healthcare funders and aid organisation to finance provision of our service to low income and marginalized communities.

One team currently has 6 members. 1 full-time and 5 part-time.

Our team has the complementary business, technical and medical expertise to take our prototype to market and is briefly profiled below.

Gift Gana, Co-Founder, is a tech-entrepreneur. He developed a bird species identification app whose image recognition technology we are adapting to medical images. Gift is also an Innovation Consultant, and has helped several European clients get millions of euros in funding under the Horizon 2020 programme.

Isak Vorster is a Radiologist. He has previous experience in the business world as Medical Science Liaison at Boehringer- Ingelheim Pharmaceuticals. He has a keen interest in the application of artificial intelligence in Radiology.

Tatenda Madzorera, Co-founder, is a Radiographer with 10 years experience. She has expertise in medical imaging quality control and assurance and health informatics. She is a 2015 recipient of The International Union of Pure and Applied Physics scientist grant.

Haji Mupakura is a seasoned Big Data and Machine Learning Engineer. He is a pioneering technologist with 8+ years of tech startup experience, and he has Co-Founded startups which include Endosit, Rentalgrid, TenantScreen and LeaseDrafter.

Evander Nyoni  is a mathematician and research scientist specializing in the practical implementation of machine learning algorithms in solving real-world problems in engineering, finance, health-care. He has experiance in analytics projects focused on fraud detection, churn market segmentation.

Andrew Chikomba is a Serial Entrepreneur, Founder of Gemwitts Enterprises where he brings a wealth of expertise in ICT and Renewable Energy ventures in Africa. Andrew is a 2015 Mandela Washington Fellow and a 2016 Cordes Fellow.

▪Ministry of Health and Child Care, Zimbabwe

Under the National TB programme, they will provide Dr CADx with the 20,000 medical images  for the Zimbabwe clinical trial scheduled to start in July 2020 which will validate the Dr CADx TB solution

 
▪Nigerian Institute of Medical Research, Nigeria

They conducted the pilot study producing a 96% accuracy of Dr CADx with the Dr CADx team using GeneXpert gold standard results for the trial.

▪StopTB program under the United Nations Office for Project Services

They provided a TB medical images dataset which we used in testing the Dr CADx prototype.

Dr CADx is targeted towards hospitals and medical imaging centres with imaging equipment who wish to improve their diagnostic accuracy and hence the quality of patient care. Hospitals and clinics in Africa both in the private and public sector generally suffer from lack of enough radiologists to review all medical images. In public hospitals and clinics imaging  cases are rarely referred to radiologists whilst in private hospitals only about 20% of imaging cases are referred to radiologists making our solution a necessity.

Distribution in international markets will be through third party medical equipment suppliers located in the targeted geographic markets and medical equipment manufacturers. Dr CADx will work leverage on the existing customer relationships that these already have which can speed up customer adoption and avoid setting up an expensive international sales team. Dr CADx will provide the necessary training to these commercial partners to ensure effective sales and servicing of its products. We will offer sales commissions of 30%.

The company will adopt a simple and scalable enterprise software as a service (SaaS) business model. With our subscription model we evaluate the customers imaging volumes in each period. We will then enter in to a monthly or annual subscription contract that gives us recurring revenue.