Computer-Aided Detection for Tuberculosis (CAD4TB)
Delft’s innovative CAD4TB software
CAD4TB was designed to help (non-expert) readers detect tuberculosis more accurately and cost-effectively. The speed of digital X-rays combined with deep learning and remote expertise make CAD4TB a valuable asset in the fight against TB.
Most countries where TB is widespread are poor; its prevalence is highest in countries with the lowest per-capita GDP. It is, therefore, not surprising that one of the main reasons TB incidences have not significantly decreased is the lack of simple, effective and inexpensive ways to detect new TB cases at an early stage. While the only way to diagnose TB is with a test like GeneXpert, these tests are expensive.
The World Health Organization
The World Health Organization (WHO) has recommended digital chest X-rays for systematic TB screening, which can show abnormalities in the lungs that suggest a patient may have TB. Although X-rays require local expertise for accurate reading, they have low operational costs and quick results.
CAD4TB Imaging Software
Software to Detect TB
Delft Imaging Systems has solved the TB detection challenge by combining low-cost, quick digital X-rays with deep learning and remote expertise. It’s called CAD4TB, Computer-Aided Detection for Tuberculosis.
If CAD4TB indicates that TB is likely present in a patient, a follow-up test like GeneXpert can be administered. This way, a patient can be diagnosed and treated within one day.
Diagnosis and treatment within one day.
Deep learning for TB detection
The CAD4TB software starts with a quick inspection of a new X-ray image. It normalizes the image to correct for input from various modalities, and then detects the many different abnormal structures that may be related to TB. The system has been trained in the detection of these abnormalities by applying deep learning to thousands of healthy and diseased X-rays from over 15 countries. CAD4TB also received expert feedback from lung specialists, making it extremely accurate and reliable.
Active TB Screening
The output of the CAD4TB software is a score between 0 and 100, indicating the likelihood that the subject on the image has active TB. The score also includes a texture heat map overlaying the original X-ray image which indicates abnormalities. Only if the output score is higher than a certain threshold will the patient need a subsequent sputum test to finalize the diagnosis. This process brings down both the cost and the time-to-treatment.
Within 30 seconds, the CAD4TB software analyses a chest X-ray, detects abnormalities and indicates the likelihood that active TB is present. This allows costly TB tests to be used only in suspected cases.