Researchers based at University of Central Florida developed the system by teaching a computer platform the optimal way to detect small specks of lung cancer in computerized tomography (CT) scans. These are of the type, through size and appearance, that radiologists sometimes have difficultly in identifying. In trials, the healthcare artificial intelligence system was found to be 95 percent accurate in total. Moreover, this was ahead of the typical scores achieved by human medics, which typically fall within the range of 65 percent accuracy.
The method used to train the artificial intelligence platform was not dissimilar to the way algorithms that facial-recognition software is taught key characteristics in relation to image analysis. To train the platform, the researchers provided in excess of 1,000 CT scans (taken from the U.S. National Institutes of Health database) to the software.
Over time the platform was taught to ignore other tissue, nerves and masses found in the CT scan images and instead to only focus on lung tissues and abnormal formations that could be tumors. The platform began to show success, and learnt to differentiate between cancerous and benign tumors. Given that successful diagnosis and treatment of lung cancer is highly dependent on early detection of lung nodules, developing a system to assist with this can help to boost patient survival rates.
READ MORE: Study reveals hybrid pythons could be new threat to Everglades
Discussing how the platform was developed, researcher Rodney LaLonde explains: “We used the brain as a model to create our system…You know how connections between neurons in the brain strengthen during development and learn? We used that blueprint, if you will, to help our system understand how to look for patterns in the CT scans and teach itself how to find these tiny tumors.”
The new medical imaging research will be presented to MICCAI 2018 (21st International Conference on Medical Image Computing and Computer Assisted Intervention), which takes place in Granada, Spain during September 2018. The associated conference paper is titled “S4ND: Single-Shot Single-Scale Lung Nodule Detection.”
