Intel AI Whitepaper

Deep Learning for Lung Cancer Detection

Kiran Vaidhya, Adarsh Raj, Krishna Chaitanya, Abhijith Chunduru, Suthirth Vaidya, Dr Ramanathan Sethuraman, Madhu Kumar, Dmitry Rizshkov, Prashant Shah

Evidence and Research

With an annual incidence of 3.34 million, lung cancer is the deadliest cancer, with anestimated 1.88 million deaths per year worldwide1. Early detection is critical towardslong-term survival; stage 4 lung cancer has a 5-year survival rate of 5% in comparisonwith stage 1 lung cancer that has a 5-year survival rate of 56%2. The National LungScreening Trial (NLST) revealed that participants who received low-dose helical CT(computed tomography) scans had a 20 percent lower risk of dying from lung cancerthan participants who received standard chest X-rays3. Advances in multi-detector CTscanning have made high-resolution volumetric imaging possible in a single breathhold, at acceptable levels of radiation exposure4. Several observational studies haveshown that a low-dose helical CT scan of the lung detects more nodules and lungcancers, including early-stage cancers5. Potentially malignant lung nodules can beidentified from chest CT scans, and early intervention can result in a higher chance oflong-term survival6.A typical chest CT scan contains anywhere in the range of 300-500 slices, and aradiologist must examine each slice to detect lung nodules. Lung nodules are smallmasses of tissue in the lung that appear as round, white shadows on a CT scan and areoften difficult to detect and document. Most are benign7, but their detection requiresspecialized expertise and with widespread implementation of lung cancer screeningprograms, the burden on radiologists is rapidly increasing.Computer-aided-detection (CAD) isbecoming increasingly useful in helpingradiologists interpret high-dimensionalimaging data like CT and MRI scans.CAD algorithms have also shown to besuccessful in increasing radiologists’ability to detect lung nodules8 9. With theadvent of deep learning and convolutionalneural networks (CNNs), CAD algorithmshave started moving away from a relianceon hand-crafted features requiringcustom engineering, to learning featuresfrom data through CNNs10 11.This whitepaper will detail how Predible Health’s deep learning algorithm for detectinglung nodules on CT scans has been optimized on Intel® Xeon® Scalable processorsusing the Intel® Distribution of OpenVINO™ Toolkit.

Read the original post
Schedule demo
Take a look at related news on this category...