AiAi.care volunteer project is working to reduce Tuberculosis, COVID-19 πŸ‘Ύ and Lung Cancer screening time and screening costs by teaching computers to "see" and interpret chest X-rays how a human Radiologist would.

We are using 700,000 labeled chest X-Rays dataset + Deep Learning to build an FDA πŸ’Š approved, open-source screening tool for Tuberculosis, COVID-19, and Lung Cancer. After an MRMC clinical trial, AiAi.care CAD will be distributed for free to emerging nations and charitable hospitals everywhere 🌏

Tuberculosis Screening

One in four people are exposed to M. Tuberculosis bacterium, but it does not become active TB unless mixed with malnutrition and overcrowding. These two factors have earned TB the nickname "disease of poverty".

Emerging nations have 12x fewer Radiologists compared to developed world, so TB patients often remain undetected while continuing to spread the bacterium further through air (coughing, sneezing, spitting). In recent years Tuberculosis is massively resurgent with 8.6 million new cases of active TB diagnosed worldwide in 2012. India accounted for a record 2.76 million new cases in 2016. A lot of these cases are MDR (Multi-Drug Resistant) TB strain.

AiAi's free TB screening tool will help emerging nations overcome shortage of Radiologists by screening X-rays within 45 seconds of capture. Early results show that our algo can potentially deliver expert-panel grade TB screening capabilities to underserved regions.

COVID-19: Algorithm in Progress

We are re-training our algorithm to screen for COVID-19 pneumonia. Subscribe to our Twitter feed for the latest updates!

Lung Cancer Screening

When it comes to cancer, early detection delivers a huge delta in survival rates compared to late detection. Unfortunately, Lung Cancer is mostly asymptomatic in earlier stages so detections in developing countries happen around Stage (III A). This late detection causes more people to die of lung cancer than of colon, breast, and prostate cancers combined.

Cancer Stage:


5-Yr Survival:


71% of lung cancers detected in chest X-Rays were visible in retrospect on previous imaging studies. Furthermore, a 1999 NIH long-term study of American Radiologists found that 19% missed lung cancers present in current chest X-Rays. These numbers may be more stark for developing nations where X-Rays are read by Primary Care Physians (PCP) instead of Radiologists.

Herein lies a 5X life-saving opportunity for early detection: We propose that a Machine-Learning screening tool with class leading sensitivity and specificity can help reduce missed-diagnose opportunities, and as a result improve survival rates 5X through early detection.

Baseline Sensitivity and Specificity Targets

- Experienced Radiologist’s Lung Cancer sensitivity is 67.5% and specificity is 91% according to (Ref: PLoS One | DOI:10.1371/journal.pone.0136624)
- Commercial CAD algorithm "Riverain OnGuard v5.2" for lung cancer detection scored 77.78% sensitivity with 32% false-positives and 55.7% test efficiency (Ref: Journal of Digital Imaging (2013) 26:651–656 DOI 10.1007/s10278-012-9565-4).
- CAD4TB by Delft Imaging scored sensitivity of 47% (95% CI range: 40-54) with 94% specificity (95% CI range: 91-97) for Tuberculosis screening (Ref: PLoS One | 2014;9:e106381).

We are aiming to beat commercial CAD algorithms by applying latest breakthroughs in Artificial Intelligence, Deep Learning, ensembling, and data augmentation.

Price of Commercial πŸ’Έ CADs

An FDA-approved commercial CADe / CADx package costs $50,000 per year for low volume radiology facilities that consult <20 patients per day. A typical large hospital license with hundreds of studies per day costs $500,000🧐 per year or more.

These costs are entirely out of reach for developing countries and charitable hospitals. This is why we are building AiAi to provide a free πŸ—½ FDA-approved, open-source alternative to the world.

CAD algorithms from GE, Philips, Siemens are too expensive for charitable hospitals in developing countries.

Donate πŸ’ your Expertise

If you are a healthcare lawyer, you can contribute your advice to our Legal Wiki here on how to donate an open-source radiology computer-aided detection algorithm for clinical use around the world.

I am a Lawyer:

You can donate your time to draft legal strategy for a unique legal challenge: donating a medical protocol, globally. Please contribute to the wiki for legal strategy by clicking above.

If you are a Radiologist, please donate one weekend to our project and help us validate the accuracy of AiAi in a MRMC study here.

I am a πŸŽ– Radiologist:

You can donate one weekend to validate AiAi CAD algorithm and its results. Please click above to fill out Radiologist volunteer form.

If you have experience in data science, machine learning, or deep learning algorithms for images, speech, or NLP text, then please help project AiAi by contributing some code / pull requests on our Github page here.

I am a Machine Learning/Deep Learning Scientist:

Dip your toes in Swish, GELU, Capsules, Multiscale-Networks, DenseNet, Wide ResNet, ResNext, PyTorch, Ensembling, XGBoost, Scikit-Image, Python 3.6, Docker, NVidia CUDA cuDNN, DICOM, HL7/FHIR, and soon PACS/VNA. Visit our Github!

If you represent a charitable hospital or non-governmental aid organization, then please contact us and we will set you up with free clinical software for computer aided detection from Digital X-ray Radiology modality.

I represent an NGO or Hospital:

If your NGO or charitable hospital can benefit from a Tuberculosis or Lung Cancer CAD, please click here to send me an email. I will set you up with free access to a cloud-hosted CAD.