DrTB

Unlock the potential for a better life with early diagnosis of TB

Comment

Inspiration

Background Tuberculosis (TB) is a chronic lung disease that occurs due to bacterial infection and is one of the top 10 leading causes of death in low resource countries (LRC).Accurate and early detection of TB is very important, otherwise, it could be life-threatening.

Current Statistics Top killers in Africa are lower respiratory tract infections, notably pneumonia, influenza, bronchitis, and tuberculosis (TB). These viral or bacterial infections of the lungs were the number two cause of death in Sub-Saharan Africa as recently as 2012, accounting for about 11.5% of total fatalities. TB (which is often counted separately) accounted for another 2.5%.

Current Problem

  1. Dearth of trained professionals : There is a lack of trained radiologists in the low resource countries (LRC), especially in the rural areas. Thus, computer aided diagnosis (CAD) systems can play an important role in the mass screening of pulmonary TB by analysing the chest X-ray images
  2. Time consuming and subjective : Chest radiographs are examined by experienced physicians for the detection of TB, however, there are subjective inconsistencies in disease diagnosis from radiograph.
  3. Misclassification : Chest X-rays (CXR) are commonly used for detection and screening of pulmonary tuberculosis. CXR images of tuberculosis are often misclassified to other diseases of similar radiologic patterns which may lead to wrong medication to the patients and thereby worsening the health condition.

What it does

Our goal is to detect whether a given person is infected with Tuberculosis or not using Artificial Intelligence by analyzing the patient's chest X-ray. To achieve this, we have built a solution using the Microsoft Azure Custom Vision API to detect TB from the chest X-ray images

How we built it

We made use of the Microsoft Azure Custom Vision API to detect whether a person was infected with TB or not by looking at their chest X-ray. Azure Custom Vision is an image recognition service that lets you build, deploy, and improve your own image identifiers. An image identifier applies labels (which represent classes or objects) to images, according to their visual characteristics. Once we build the model we exported and rendered the results using Javascript, HTML and CSS.

Dataset used to train the model

To train the Azure image recognition classifier, we used publicly available datasets comprising 3500 TB infected and 3500 normal chest X-ray images.

Challenges we ran into

The training image data was vey large and it took us a lot of time to build the model and deploy it.

Accomplishments that we're proud of

We were able to achieve a high precision(99%) and recall(99%). Our model performs very well in differentiating the TB X-ray images from the non-TB X-ray images. We tested our model's performance on unseen X-ray images(test data ) and were accurately able to detect the presence of TB.

What we learned

We learnt how to build and deploy a ML model using the Microsoft Azure Custom Vision API .

What's next for DrTB

We plan to go one step further and enhance the performance of our model by segmenting the image to identify the area of the disease within the chest X-ray and further performing a deeper analysis to identify the severity and type of infection.

Built With

Share this project:

Updates