Inclusive Medical Imaging

To provide a medical imaging solutions that has been built for Africans by Africans

Comment

Inspiration

One of the hidden factors to financial wellbeing is a person's health. Healthcare in Africa is in a sorry state and this solution aims to help improve healthcare in Africa. Many people in Africa and misdiagnosed or given late diagnosis and end up with high medical bills or even death due to this. The loss of a family's breadwinner can lead to a family plunging into poverty or sometimes even draining all the family's resources through paying of medical bills.

We seek to reduce the time taken to diagnose medical conditions with the use of image classification by using chest X-rays and CT-Scans to help in the detection of conditions such as Pneumothorax, Pleural Effusion, TB and Intracranial Hemorrhage through Augmented Intelligence (AI).

We seek to add a Triage solution to help prioritize emergency cases within a busy hospital environment and during high load emergency cases to eliminate the traditional workflow scan of First in First out to priority driven approach. We also seek to increase the accuracy and level of confidence of a diagnosis by adding in an extra layer of screen to medical images. Most of the solutions available are trained on data from foreign countries. The genetic makeup, disease preference and body(organ) structure differing between races, there is greater need to have a solution that is within to improve both the specificity and positivity of the said solution.

What it does

Using deep learning methods to develop a pre-trained model/s that can be used in the detection of medical conditions such as TB and later on Pneumothorax, Pleural effusion among others using Medical Images (x-rays, mammograms and CT scans). Our solution will be done by Kenyans for Kenyans, and hence increasing its effectiveness within the Kenyan (and African) space.

How we built it

The solution will use Convolutional Neural Networks (CNNs) and Deep Learning to detect patterns in medical images and learn from the patterns to accurately classify the images. We also seek to have an AI (Augmented Intelligence) system that fits seamlessly with the existing Hospital IT infrastructure to ensure ease of adoption. After an image is taken and sent to our system, it is passed through the trained model and classified over the various conditions. If the model predicts a positive for any of the conditions, it flags the image for further analysis by the doctor and immediately deletes the patient’s data automatically. With the use of APIs (Application Programming Interface), our system will easily integrate with multiple hospital systems

Challenges we ran into

Given the high complexity of running deep learning models, it took a lot of time to create the model and ensure it works effectively. We also had a challenge of acquiring data to use to train the model as medical data is very sensitive and hospitals are reluctant to release the data

Accomplishments that we're proud of

We were able to train a model to high accuracy of about 92% which can be further improved on by fine tuning the deep learning model.
We hope that this solution will create access to proper healthcare and reduce cases of misdiagnosis and late diagnosis and hence save millions of families mental, emotional and financial anguish.

What we learned

  • An estimated 417,000 people died from the disease in the African region (1.7 million globally) in 2016. Over 25% of TB deaths occur in the African Region. These are unnecessary deaths of breadwinners and labour force that could be used to gain wealth.
  • 3 in 10 Kenyan Patients Get Misdiagnosed, Leading to Death or Permanent Disability. Available here

What's next for Inclusive Medical Imaging

  • Radiotherapy solution for contouring organs at Risk (OAR) to enable fast planning within a minute. The system will get a CT simulation scan and contours (annotate) the image and sends it back for treatment planning. It intends to configure once and implement to the workflow forever.
  • Venturing in to the diagnosis of other non-imaging diseases using machine learning models to improve the healthcare space in Africa.

Built With

Share this project:

Updates