Artificial intelligence can save lives if used correctly. This theory was recently tested by a couple of engineers and medical experts when they came together to create an AI system that is able to detect cancer tumors that are often missed by radiologists. If used in real life, this technology will be able to save scores of lives every year and boost patient survival rates.
This initiative to explore an AI system’s ability to save lives was taken by researchers who are currently based at the University of Central Florida. They created a system by teaching a computer platform the best way to spot specks of lung cancers visible in CT scans that are often too small or have a different appearance to be highlighted as cancer by radiologists. During the trials, the AI system had 95 percent accuracy. The results were way ahead of the scores achieved by human medical professionals, i.e., 65 percent. So, the AI system was approximately 30 percent more accurate as compared to humans.
During the research, the artificial intelligence platform was trained by using a method that’s very like the way algorithms are used by most of the facial-recognition software to learn key characteristics with regard to image analysis. To offer training to the AI platform, more than 1,000 CT scans accessed from the National Institutes of Health database were provided to it.
Over a considerable span of time, the AI platform was taught to focus on abnormal formations on lung tissues that can potentially be tumors while ignoring other tissue, nerves or masses present in a CT scan. The platform began showing signs of success when it learned to differentiate between benign and cancerous tumors.
It is a well-known fact that successful treatment of lung cancer depends largely on how soon it is detected. So, developing this AI system would surely save lives and increase patient survival rates.
Using Brain as the Model
One of the members of the research team, Rodney LaLonde stated that the researchers used the brain as a model to create the AI system. The connections between different neurons in the brain usually strengthen during the process of learning or development. The researchers used this blueprint to help the Ai system understand how to find patterns in CT scans and teach itself the process of finding the miniscule tumors.
This research will be presented at 21st International Conference on Medical Image Computing and Computer-Assisted Intervention, also known as MICCAI 2018 that will take place in September 2018 in Spain. The conference paper is titled as S4ND: Single-Shot Single-Scale Lung Nodule Detection.