Reading chest x-rays is one of the most complex radiology tasks and has high inter-reader variability. Compared to other methods, chest x-rays are popular due to lower doses of radiation, lower cost, and reduced time. As a result, chest x-rays are widely used as a screening tool to provide a low fidelity view which paves the way for more sophisticated imaging methods. HCL Technologies, a member of the Intel® AI Builders program, has developed a solution that leverages artificial intelligence technologies from Intel to help detect patient abnormalities in chest x-rays.
Chest x-ray diagnosing requires not just expertise, but also availability of radiologists
With approximately 2 billion procedures completed per year, chest x-rays are the most common imaging examination tool used in medical practice and are critical for screening, diagnosis, and management of a variety of conditions, including pneumonia. Although x-rays are extremely helpful as an imaging tool, detecting pneumonia in chest x-rays is a challenging task that relies on the expertise of radiologists. The application of AI in the initial x-ray image screening helps to make the process of diagnosis more efficient for radiologists and reduce errors.
Intel® Distribution of OpenVINO™ toolkit is a key differentiator for inferencing in medical imaging
To train their model, HCL used CheXNet, a 121-layer convolutional neural network, on a National Institute of Health (NIH) dataset that contains 112,120 frontal-view X-ray images of 30,000 unique patients. The images were pre-processed with the CLAHE (Contrast Limited Adaptive Histogram Equalization) technique. By leveraging Horovod, a distributed training framework, HCL was able to reduce model training time by ~4x per epoch. HCL performed their model training on an Intel Xeon® Platinum 8153 Processor-based system. The trained model was optimized using the Intel Distribution of OpenVINO toolkit, where improvements in inference and throughput time were observed. The improvement in inference and throughput time was also observed when inferred on Intel Xeon Platinum 8256 Processor (3.80 GHz) and the Intel Neural Compute Stick 2 processor.
About HCL Technologies Ltd.
HCL Technologies is a next-generation global technology company that helps enterprises reimagine their businesses for the digital age. Our technology products, services, and engineering are built on four decades of innovation, with a world-renowned management philosophy, a strong culture of invention and risk-taking, and a relentless focus on customer relationships. We offer an integrated portfolio of products, solutions, services, and IP through our Mode 1-2-3 strategy, built around Digital, IoT, Cloud, Automation, Cybersecurity, Analytics, Infrastructure Management and Engineering Services, amongst others. With a worldwide network of R&D, innovation labs and delivery centers, and 137,000+ ‘Ideapreneurs’ working in 44 countries, HCL serves leading enterprises across key industries, including 250 of the Fortune 500 and 650 of the Global 2000.
Learn more about HCL’s optimized market-ready solutions in the Intel AI Builders Solutions Catalog
 Raoof, S., Feigin, D., Sung, A., Raoof, S., Irugulpati, L., & Rosenow, E. C. (2012). Interpretation of plain chest roentgenogram. Chest, 141(2), 545-558. https://doi.org/10.1378/chest.10-1302