AI System to Generates Synthetic Scans of Brain

These scans are used to teach AI diagnostic system
18 September 2018   885

A group of researchers developed an artificial intelligence capable of generating sets of images of an MRI of a human brain. The technology is designed to increase the effectiveness of training AI, specializing in the diagnosis of brain cancer. Tests showed that the effectiveness of diagnostic programs trained on generated kits increased by 14%.

The project was implemented jointly by specialists from NVIDIA, the Mayo Clinic and the Clinical Data Research Center. Development based on the generative and adversarial network structure (GAN) was conducted on the NVIDIA DGX platform using the PyTorch deep training systems. Two interconnected artificial intellects were used. One network generated its own MRI snapshots on the basis of real ones, and the second tried to distinguish real from fake ones.

GAN automatically marks the created sets of MRI images, which significantly speeds up learning. With manual annotation, this work takes experts many hours. In addition, since the system does not consider the brain and tumor as a whole, the operator can correct the picture by moving the tumor or changing its size.

Hu Chang, one of the authors of the study, said that the generated MRI kits also solve the problem of using confidential information. These pictures form a medical secret, and permission is required to use them. And the resulting system can be publicly available.

Hardware limitations forced the team to reduce the resolution of the original images by 8 times. Also, at the moment, neoplasms sometimes look "superimposed" on a snapshot. In the future, researchers plan to eliminate these shortcomings.

When teaching neural networks-diagnosticians, the question of the availability of training datasets is relevant. Developed by German scientists, AI, which determines myocardial infarction by ECG, used as input only 200 records. According to the creators, this seriously worsened the efficiency of the system. Tools that create datasets for learning neural networks are designed to help solve this problem.

Nvidia to Open SPADE Source Code

SPADE machine learning system creates realistic landscapes based on rough human sketches
15 April 2019   858

NVIDIA has released the source code for the SPADE machine learning system (GauGAN), which allows for the synthesis of realistic landscapes based on rough sketches, as well as training models associated with the project. The system was demonstrated in March at the GTC 2019 conference, but the code was published only yesterday. The developments are open under the non-free license CC BY-NC-SA 4.0 (Creative Commons Attribution-NonCommercial-ShareAlike 4.0), allowing use only for non-commercial purposes. The code is written in Python using the PyTorch framework.

Sketches are drawn up in the form of a segmented map that determines the placement of exemplary objects on the scene. The nature of the generated objects is set using color labels. For example, a blue fill turns into sky, blue into water, dark green into trees, light green into grass, light brown into stones, dark brown into mountains, gray into snow, a brown line into a road, and a blue line into the river. Additionally, based on the choice of reference images, the overall style of the composition and the time of day are determined. The proposed tool for creating virtual worlds can be useful to a wide range of specialists, from architects and urban planners to game developers and landscape designers.

Objects are synthesized by a generative-adversarial neural network (GAN), which, based on a schematic segmented map, creates realistic images by borrowing parts from a model previously trained on several million photographs. In contrast to the previously developed systems of image synthesis, the proposed method is based on the use of adaptive spatial transformation followed by transformation based on machine learning. Processing a segmented map instead of semantic markup allows you to achieve an exact match of the result and control the style.

To achieve realism, two competing neural networks are used: the generator and the discriminator (Discriminator). The generator generates images based on mixing elements of real photos, and the discriminator identifies possible deviations from real images. As a result, a feedback is formed, on the basis of which the generator begins to assemble more and more qualitative samples, until the discriminator ceases to distinguish them from the real ones.