The Doctor Alzimov system can assess
patient CT scans for the presence of
lung cancer in less than 20 seconds

Russian researchers develop new lung cancer detection software

February 07, 2019
by John R. Fischer, Senior Reporter
Russian researchers have developed a new intelligent software system that can assess the presence of lung cancer within mere seconds.

Named Doctor Alzimov after science-fiction writer Isaac Asimov, the solution is designed to assess CT scans within 20 seconds with artificial intelligence and produce an image in which the pathology is clearly marked.

"Lung cancer is among the most widespread diseases in the world, including the Russian Federation. In Russia, there are not so many clinics that are engaged in developing the systems for lung cancer diagnostics, so this inspired our scientific group to start our research," Professor Lev Utkin, head of the Research Laboratory of Neural Network Technologies and Artificial Intelligence of Peter the Great St. Petersburg Polytechnic University (SPbPU). "So when we discussed the area of research with the doctors from St. Petersburg Clinical Research for Specialized Types of Medical Care (Oncological), we decided that this area is the most prospective for the use of the artificially intelligent system."

Clinicians can spend 30 minutes to hours evaluating the CT of one patient. Doctor Alzimov acts as an assistant to the doctor by speeding up the process of analysis.

Initially setting up its algorithm to search for nodules starting from six millimeters, the size of tumors that radiologists start treatment for, the researchers found the system to be capable of detecting nodules of even smaller sizes.

The purpose of the solution is to differentiate malignant and benign nodules as well as metastases. It can also detect metastasizes from different organs and non-oncological diseases, such as tuberculosis.

The system relies on the chord method, a newly proposed and developed approach for lung cancer classification that was patented within only three months. Using segmented CT images, the technology randomly draws points on the surface of a nodule, with chords connecting them. The length histogram of the chords reflects the shape and structure of the tumor.

To learn more about external surroundings of each nodule, the tumor is placed in a cube with perpendiculars drawn from its edges to the surface of the nodule. This creates a compact and simple histogram for the nodule that can be read by the Doctor Alzimov system, rather than a graphically complex and heavy image of the CT.

Training for the system was derived from the analysis of 1000 CT images from LUNA 16 and LIDC datasets, as well as images of about 250 patients whose information was stored in the researchers’ own data set, Lung Intelligence Resource Annotated.

The assessment of each new image enables the system to self-improve, with researchers relying on the supercomputer center, Polytechnic, to speed up learning and testing processes. They plan to reduce diagnostic testing time per patient from 20 to 2 seconds in the future by transferring images to the supercomputer using the internet, with the radiologist receiving the marked image, rather than the large CT image. This decreases time for analysis and diagnostics.

"The system is not only an artificial intelligence system, it is a whole digital platform. Using the cloud service every clinic will be able to connect to the platform to analyze the CT of their patients," said Utkin. "This data will be transferred to supercomputer "Polytechnic". So the hospitals don't need to install the specific software, but just to connect the digital platform.".

The researchers plan to increase the number of images by four times by mid-2019 and at some point in the future, will adapt the system to assess results of ultrasound and X-ray in the evaluation of other organs.

Open testing of the system will take place at the beginning of 2019, with the system first used at the St. Petersburg Clinical Research Center for Specialized Types of Medical Care (Oncological) and eventually extended to other institutions.

The research team consisted of staff from Peter the Great St. Petersburg Polytechnic University (SPbPU) and the Russian Academic Excellence Initiative participant, and radiologists from St. Petersburg Clinical Research for Specialized Types of Medical Care (Oncological).

The project was supported by the Russian Science Foundation.