Our projects

• 

  Shape Quality Checking

  By Dmitry Petrov, Egor Kuznetsov, Mikhail Belyaev | Status: Active

• 

  Heritability in Connectomes

  By Anvar Kurmukov, Boris Gutman, Yulia Denisova | Status: Active

• 

  Metric Learning via Diffeomorphic Image Registration

  By Ayagoz Mussabayeva, Boris Gutman, Anvar Kurmukov, Yulia Denisova | Status: Active

  Large Deformation Diffeomorphic Metric Mapping method induces a Riemannian structure on the space of diffeomorphism between images. The main goal is to optimize the parameters of the Riemannian metric to build a representative geometry. In practice, this may lead to a biology-aware registration, focusing its attention on the predictive task at hand such as identifying the effects of a disease.

• 

  MRI to CT Synthesis

  By Maxim Pisov, Artem Pimkin, Vladimir Kondratenko, Mikhail Belyaev | Status: Active

  Both MRI and CT brain images are required for diagnostics and treatment. Unfortunately obtaining up-to-date MRI and CT is time-consuming and expensive or even impossible due to contradictions.

  That's why synthesizing a corresponding CT image to an exisiting MRI (or vice versa) is a challenging but important task.

• 

  Dataset Driven Optimal Parcellation

  By Anvar Kurmukov, Boris Gutman, Ayagoz Mussabayeva, Yulia Denisova | Status: Finished

  Deriving connectivity-based brain atlases from individual connectomes is an open challenge. Actually, aggregating the individual parcellations into a consensus parcellation, we find that the parcellation based-atlas outperform all other parcellation-based atlases.

• 

  Brain Tumor Image Retrieval

  By Gleb Makarchuk, Maxim Pisov, Mikhail Belyaev | Status: Active

  Building an effective retrieval system for brain tumors is an open problem. Such a system will help doctors to predict the development of diseases and treatment planning based on analysis of clinically similar cases.

• 

  Ensembling Neural Networks for Histological Images

  By Gleb Makarchuk, Vladimir Kondratenko, Artem Pimkin, Maxim Pisov, Egor Krivov, Mikhail Belyaev | Status: Finished

  Adaptation of the state-of-the-art convolutional neural networks (CNN) architectures for digital pathology images analysis.  The project was developed as part of participation Breast Cancer Histology Challenge dataset and we obtained 90% accuracy for the 4-class tissue classification task.
• 

  Deep Learning for Brain Radiosurgery

  By Egor Krivov, Boris Shirokikh, Mikhail Belyaev | Status: Active

  The gamma-knife project is focused on improving brain radiosurgery procedures with modern deep learning. We have created a system, generating precise delineation of metastases on MRI. Then integrated it into the everyday pipeline of the Moscow Gamma-Knife facility, where it significantly speeds up operations and provides the second opinion for doctors. Future work will be focused on adding other types of tumors, primarily meningioma and schwannoma and providing delineation for follow-up visits, to generate a volumetric history of each tumor of the patient.

• 

  Analysis of Morphometry and EEG in Case of Schizophrenia Disorder

  By Nikolay Lutsyak, Mikhail Belyaev | Status: Closed

  Analysis of multimodal (morphometry, EEG) data from brain to study schizophrenia using ML. The effect of different atlases on classification and search for intact zones.

• 

  White Matter Hyperintensity

  By Amir Safiullin, Boris Shirokikh, Mikhail Belyaev | Status: Active

  Building an automatic segmentation algorithm of White Matter Hyperintensities (WMH) of presumed vascular origin. The project arose from the WMH segmentation challenge at MICCAI 2017, where we are able to compare our model with state-of-the-art approaches and other competitors. Apparently, it appeared to be an almost perfect "playground" to test and compare different deep learning, MRI pre-processing and medical data augmentation approaches.

• 

  DeepPipe

  By Egor Krivov, Maxim Pisov | Status: Active

  A collection of tools useful for medical image analysis, including preprocessing, data augmentation, model training, performance validation and final prediction.

• 

  Predicting Conversion of MCI to AD

  By Yaroslav Shmulev, Mikhail Belyaev | Status: Active

  The conversion prediction is an important problem as approximately 15% of patients with MCI converges to AD every year.  In the current work, we are focusing on the conversion prediction using brain Magnetic Resonance Imaging and clinical data, such as demographics, cognitive assessments, genetic, and biochemical markers.

• 

  Geometry of the Set of Symmetric Positive Semidefinite Matrices

  By Daria Belyaeva, Mikhail Belyaev, Yulia Denisova | Status: Finished

  We propose to transform structural connectomes by taking their normalized Laplacians prior to any analysis, to put them into a space of symmetric positive semidefinite (SPSD) matrices, and apply methods developed for manifold-valued data. The geometry of the SPD matrix manifold is well-known and used in many classification algorithms.

• 

  Generative Models of Connectomes

  By Ayagoz Mussabayeva, Anvar Kurmukov, Yulia Denisova | Status: Frozen

  Usually, the size of the sample in medical data is critically small. One method for dealing with small sample sizes in machine learning is artificial data replication or so-called augmentation. The "good" generative model can show the basic principles of the structure of the human brain network.

• 

  Huntington Disease

  By Artyom Borzov, Igor Medvedev, Nikolay Lutsyak, Yuri Seliverstov, Mikhail Belyaev | Status: Active

  Huntington's disease (HD) is a neurodegenerative disease caused by an increase in the number of CAG repeats in the HTT gene. Collecting clinical data for HD gene expansion carriers allow people open the new view on HD investigation: use Machine Learning Approach (MLA) to find new information about HD.

  image source

• 

  Connectome Classification

  By Dmitry Petrov, Yulia Denisova, Anvar Kurmukov, Anna Tkachev, Alexander Ivanov | Status: Finished

  The structural connectome classification is a challenging task due to a small sample size, high dimensionality of feature space and high rate of noise. Connectome classification results show structural differences in the neural network of the human brain. Using different methods was reached state-of-art results in this task.

• 

  Convolutional Networks for 3D MRI Data Analysis

  By Sergey Korolev, Amir Safiullin, Mikhail Belyaev | Status: Finished

  Modern deep Learning methods are competitive to classic ML algorithms for classification task on neurodegenerative diseases based on neuroimaging data. Also, this methods can show interpretable results with contemporary technique so-called attention. The results confirmed with well-known facts about the neurodegenerative disease.