Santhosh Subramanian

Currently In High School

NCI Radiation Oncology

Fellow

NLST (Learn more about this study)

NLST-149

Aug 3, 2015

Using Machine Learning and Artificial Intelligence to Distinguish Between Malignant and Benign Lung Cancer Tumors in a CT Scan

By using Machine Learning and Artificial Intelligence, we want to develop a novel algorithm that will be able to distinguish between Malignant and Benign Lung Cancer Tumors found in a CT Scan. In the Lung Cancer Screening Trial with 53,454 patients, about 96.4% of patients in the trial were found as false - positives. Meaning that based on the CT scan, the researchers thought that there was a malignant tumor about 96 out of 100 times, whereas the tumor was only benign, thus there were costs for unneeded invasive procedures. With the large number of data - sets found in the NSLT database, we want to harness the CT Scans to develop an algorithm that will be able to distinguish between malignant and benign lung cancer tumors. By doing this, we can reduce overall costs, relieve anxiety, improve diagnostic measures, and create a better environment for lung cancer screening. The implementation of Machine Learning and Artificial Intelligence will allow the algorithm to grow and learning and differentiate between malignant and benign tumors at a hundred percent rate. If developed, we can use this algorithm as a baseline to differentiate and detect tumors in other cancer imaging techniques such as PET, Mammography, MRI, Nuclear Medicine, Bio-luminescence, and Fluorescence imaging. This could also pave way to better implementation of predictive analysis in cancer imaging. With the use of Machine Learning and Artificial intelligence, we could better predict based on imaging techniques how the tumor will grow and where it will spread.

Computer vision and image processing algorithms will be used for feature extraction, detection / segmentation, and high level processing. A machine learning algorithm for pattern recognition will then be implemented to discern patterns of differences between images of benign nodules and malignant tumors.

1) Develop a Novel Algorithm to Distinguish Between Malignant and Benign Lung Cancer Tumors in a CT Scan

2) Use machine learning and artificial intelligence to create a diagnostic method with hundred percent accuracy.

3) To use this algorithm as a baseline to develop algorithms to distinguish and characterize tumors for different cancers and different cancer imaging techniques.