Doctor AIzimov
Polytech scientists have created an intelligent system for diagnosing tumors in the lungs
Ilona Zhabenko, SPbPU Media Center
Scientists of the Polytechnic University together with doctors of the St. Petersburg Clinical Scientific and Practical Center for Specialized Types of Medical Care (oncological) have developed an intelligent system for diagnosing tumors in the lungs.
The software, which can be installed on any computer, analyzes the computed tomography of patients' lungs in 20 seconds and issues a conclusion in the form of a visually highlighted pathology. The developers named the system Doctor AIzimov (AI – Artificial Intelligence, that is, artificial intelligence) in honor of the science fiction writer Isaac Asimov, who proclaimed three famous laws of robotics.
At the end of 2018, closed tests of the intelligent system were conducted. She analyzed anonymized CT scans of 60 new Cancer Center patients. According to doctors, the tests were successful, as the system detected focal formations in the lungs even of small sizes (2 mm).
"Initially, we set up the algorithm to search for foci from 6 millimeters, since doctors themselves adhere to this tactic. But the system is so trained that it finds even smaller neoplasms on its own," explained Lev UTKIN, project manager, head of the Research Laboratory of Neural Network Technologies and Artificial Intelligence.
The laboratory was created on the basis of the Polytechnic University, but in addition to the university staff (Lev Utkin, Mikhail Ryabinina and Alexey Lukashina), it employs specialists of the Oncological Center – the head of the department of radiation diagnostics Anna Meldo and radiologist Ivan Prokhorov. The project of a multidisciplinary team of specialists received the support of the Russian Science Foundation, and the technology of tumor detection using the chord method received a patent in a record 3 months.
The chord method consists in the fact that on the CT scan image, points are randomly placed on the surface of the tumor, which are then connected by segments – chords. The histogram of the lengths of these segments reflects the shape and structure of the tumor. This is how scientists study the neoplasm from the inside, but it is no less important what surrounds it from the outside. To do this, the tumor is conditionally placed in a cube, and perpendicular lines are drawn from its faces to the surface of the neoplasm. Thus, instead of a graphically complex and volumetric CT scan (its size can reach 1 GB), the tumor is presented in the form of compact and simple histograms, which are analyzed by Doctor AIzimov.
Before the system started working, it was necessary to teach it to identify neoplasms and distinguish malignant tumors from benign ones. "On a CT scan, you can see a lot of objects, and the main task is to teach the system to recognize what each of them is. By bringing the learning process of the system closer to the logic of a doctor and using clinical and radiological classification, we are trying to teach the system not only to detect neoplasms, but also to distinguish other diseases similar to cancer," comments Anna Meldo, head of the radiology department.
The system was trained on about 1000 CT scans from the LUNA 16 and LIDC databases. The laboratory staff is also creating their own database, the LIRA – Lung Intelligence Resource Annotated, where at the moment you can find pictures of about 250 patients. By mid-2019, it will increase fourfold.
With each new snapshot, the system improves itself. The Polytechnic supercomputer center helps her in this, which speeds up the algorithm setup. It is planned that the patient's images will be sent via the internal network to the Polytechnic supercomputer center, where their processing will take not 20, but two seconds, after which the doctor will receive a marked image and will work with it, and not with volumetric computed tomography – this reduces the analysis time.
After closed testing, an open test of the system is planned, then it will be implemented in the Cancer Center. In the future, scientists want to develop the project and involve other medical institutions in it. Over time, the algorithm can be taught to analyze ultrasound and X-rays of other internal organs. All data will be processed by a supercomputer, and the results issued by the system will be sent to doctors.
Portal "Eternal youth" http://vechnayamolodost.ru