Artificial intelligence-based software has been developed by scientists in Russia that can accurately identify lung cancer tumors from CT scans within 20 seconds. While the software is not intended to replace the trained eyes of radiologists, the software could be used to speed up diagnosis of lung cancer allowing treatment to be commenced much faster.
The software, named Doctor Alzimov, can be installed on any computer and used to analyze the results of CT scans for the signs of cancer. The software was trained using 1000 CT images from LUNA 16 and LIDC datasets, in addition to a 250-patient data set collected by the researchers. While already accurate, detection improves with each scan that the software analyses. The team plans to increase the number of training images four-fold by the middle of 2019.
Scientists at Peter the Great St.Petersburg Polytechnic University (SPbPU) completed the first full test of the software in late 2018. CT scan images from 60 patients at St.Petersburg Clinical Research for Specialized Types of Medical Care were analyzed using the software.
Within 20 seconds, each CT scan was analyzed and the software generated images from the CT scan with focal nodules clearly marked. Instead of a radiologist being provided with a full CT scan for manual analysis, much smaller images are provided which only include the marked areas where the nodules have been identified by the software.
Typically, treatment is commenced on nodules that are 6mm or greater in size. Initially the researchers trained their software to look for nodules of 6mm, but they found that the software was capable of detecting nodules as small as 2mm with a high degree of accuracy. The software has also been trained to differentiate between benign and malignant tumors.
The researchers believe that it can be used to greatly speed up the analysis of CT scans, speed diagnosis, and allow treatment to be commenced much more rapidly. The researchers also plan to use the software on the SPbPU supercomputer which they believe will speed the analysis even further, reducing the time spent analyzing each CT scan from 20 seconds to 2 seconds.
Open testing of the software is now underway, following which it will be used at the St.Petersburg Clinical Research for Specialized Types of Medical Care before being extended for use in other medical institutions. The researchers are also redeveloping the software to analyze ultrasound and X-ray images to search for and identify tumors in other organs.