Accessible Healthcare Using AI - Qure.ai & RedBrick AI

Qure.ai is an innovative artificial intelligence (AI) provider with a mission to make healthcare more accessible and affordable. The company uses deep learning to offer automated interpretation of radiology exams, including X-rays, CT scans, and ultrasound scans. This not only conserves resources for hospitals and saves time for radiologists, but also enhances the accuracy of diagnoses.

Qure AI's suite of products, AI for Chest X-ray, Chest CT, and Head CT, have been adopted by institutions in over 70 countries and their products serve over 10 million patients every year!

Improving Lung Health using AI

Qure AI's lung health products are some of the most widely adopted AI algorithms in the industry. When Qure.ai was founded in 2016, they started with a Chest X-ray product that can now detect over 30 abnormalities and provide automated reporting. Since then, Qure.ai has launched products for the early detection of lung cancer using Chest CT.

Lung cancer is responsible for 1.74 million deaths a year

Lung cancer is the leading cancer killer in both men and women in the U.S., responsible for about 1 in 5 of all cancer deaths. Each year, more people die of lung cancer than of colon, breast, and prostate cancers combined.

Early detection of cancer is instrumental in increasing chance of su rvival. From the graphic below, no tice how the 5-year survivalratefor non-small cell lung cancer drops from 65% to 9% when the cancerspreads.

Medical imaging plays an important role in diagnosing and grading lung cancer. Screening patients with CT scans, as opposed to chest X-rays, has resulted in a 20% reduction in lung cancer-specific deaths.

Empowering lung cancer screening with AI

Radiologists face several challenges in accurately diagnosing lung cancer, despite having access to the right imaging tools. For instance, according to a study, 42.5% of malpractice suits against radiologists result from a failure to diagnose lung cancer.

Radiologists often disagree on factors such as the presence of cancerous nodules, nodule texture, and nodule size, leading to a common source of error. Qure AI's products can increase inter-reader agreement, which measures how often different radiologists agree on the prognosis of a single scan, by over 33%.

Diagnosing lung cancer is not a one-time event, as the NELSON lung protocol recommends measuring the growth of nodules to assess malignancy after a patient has been diagnosed. However, measuring the volumes of multiple nodules in chest CT scans and comparing them over time can be a tedious exercise.

Qure AI's AI for Chest CT products address several real-world challenges faced by radiologists. Specifically, Qure.ai can:

To train their high-performance deep learning algorithms, Qure.ai had to amass a large annotated dataset for each use-case. The Qure.ai team uses the RedBrick AI platform to annotate scans, manage projects, and enforce quality standards.

Purpose built annotation for Radiology AI

Qure.ai's products, ranging from 2D X-ray reporting to 3D nodule detection in Chest CT, require highly specialized annotation and workflow tools.

Before transitioning to RedBrick AI, Qure.ai used a combination of internal and open-source tools, such as 3D Slicer. However, the Qure.ai team found that they were spending a significant amount of time maintaining their internal labeling infrastructure. By transitioning to a platform like RedBrick AI, they can now better focus on model building.

After switching to RedBrick AI, we experienced several benefits:

• Onboarding and training annotators became much easier.
• RedBrick AI's review workflows improved project management.
• The time it takes to load a scan for annotation was drastically reduced.
• RedBrick AI's taxonomies allowed us to capture more structured data.

Prakash Vanapalli, Director of Data Science

Quality control with consensus and validation

As discussed above, the diagnosis and classification of cancerous lung nodules can be subjective. However, to train an accurate algorithm, annotations representative of the truth are needed, rather than a single annotator's opinion. Qure.ai uses RedBrick AI's consensus tools to arbitrate between several annotator opinions.

Radiology annotations can be subjective. For example, radiologists may often disagree on the size of a nodule in a chest CT scan. Therefore, we need excellent tools to help us get closer to the truth. RedBrick AI allows us to give a single CT scan to multiple radiologists and view all of their opinions overlapping, which is incredibly insightful. With RedBrick AI, we can manage the entire workflow of measuring consensus during our annotation projects.

Souvik Mandal, Senior AI Scientist

Arbitrate between multiple annotations and measure consensus

Sometimes disagreement between annotators may result in lower quality annotations. However, not adhering to the labeling schema can also be a big source of annotation quality problems. For example, forgetting to annotate a particular category or attribute. These "schema" mistakes are usually found at the end of an annotation project, when engineers are preparing to train with the data.

Previously, we used to perform extensive post-processing of our annotations in order to identify errors such as missing label attributes. Now, we use RedBrick AI's custom label validation script to automatically check for these mistakes in real-time. This has significantly reduced the post-processing burden and saved us a considerable amount of time.

Ritvik Jain, Product Manager

Write a custom JS script to validate labels

Clinical performance for regulatory approval

In 2023 the Qure.ai team received CE Class IIb certification for their Chest CT product. To receive regulatory clearance, you must conduct a clinical performance study, where you measure the performance of the algorithm against a reference standard. Qure.ai used RedBrick AI in collaboration with their European partners to conduct this clinical performance study.

While pursuing Class IIb certification, being able to deploy RedBrick AI within a research institutions private environment was crucial. Doing so restricted access to data within a private environment building a lot of trust in the system.

Ritvik Jain, Product Manager

Qure.ai's AI products for radiology are improving healthcare accessibility and affordability while saving lives. RedBrick AI is excited to help Qure.ai build high-performance deep learning algorithms with our annotation and workflow tools!

The best thing about RedBrick AI is their excellent support. They truly listen to our feedback and act on it quickly. I have worked with their competitors before, and this level of support really sets them apart. In fact, they really stand out compared to any company we've worked with.

Ritvik Jain, Product Manager