Сellular pathology detection
Goal
Develop a solution of a medical diagnostic task based on neural networks with a team of schoolchildren within a 3-week shift in the educational center ‘Sirius’. The service should accurately detect the presence of cellular anomalies by micrograph of a living tissue section.
Our solution
During preparation, more than 1000 photographs of buccal epithelium samples were taken.
An application was developed In order to artificially increase the number of photos by rotating and adding noise.
The total size of the photo database is 22 000 images.
Photo marking was done manually with the subsequent use of regular expressions. Four main cell types were distinguished:
Normal
cell
Binucleated
cell
Cell with a
micronucleus
Cell with a
protrusion
The solution allowed to accurately determine the percentage of cells with anomalies.
The results for people with different smoking experience:
Smoking experience: 0 years
Smoking experience: 20 years
The solution will allow to:
-
Minimize the human factor influence
-
Speed up image studying by more than 300 times
-
Get rid of a large amount of documentation for each study
-
Standardize the process of studying cellular anomalies
The team involved in the service development
Leaders:
Bazarnyi Vladimir Viktorovich (doctor)
Maksimova Arina Yurievna (MLS)
Consultant:
Alperin Yakov Sergeevich (IT Specialist)
Team of students:
Ershova Ksenia
Photo layout, dataset creation
Tonkoshkurov Nikita
Neural network training,
web server creation
Kiseleva Ekaterina
Photo archive creation
Sarapulova Arina
Photo archive creation
Shcherbakova Maria
Photo archive creation
Semenov Arseny
Neural network training,
web server creation
Chermnykh Dmitry
Neural network training,
web server creation