MedTech

keys 3

Global Ai integrator

1. Analysis of doctors' diagnoses and mammography

Task:

Development of a recommendation system that issues a diagnosis based on the data filled in about each patient

Decision:

A recommendation system offering its own variants of conclusions based on the input features that are filled in during the formalized descriptions of each specific study

Result:

Improvement of the accuracy of differential diagnostics in the process of daily practical activities of radiation diagnosticians

Program analysis of mammography

Automatic image analysis systems. An autonomous analysis of images obtained in the course of various studies by methods of radiation diagnostics is carried out. Based on the analysis carried out, the systems offer their own variants of conclusions for each specific study.

2. AI determining the diagnosis of lung disease by X-ray

Task:

Based on the input images of X-ray images of the human chest, determine the diagnosis with an accuracy of at least 80%, identify the area of pathology with an accuracy of at least 70%

Result:

Increasing the capacity of the medical facility

Increasing the accuracy of diagnosis

Solution:

A recommendation system offering its own variants of conclusions based on the input features that are filled in during the formalized descriptions of each specific studyStep 1. Detection (A retina net-based detector is used, with which we will find objects such as wires, supports, garlands of insulators, vibration dampers)

Possibility of remote diagnosis (implementation of telemedicine functions)

3. AI recommendation systems for the diagnosis of diseases (COVID-19)

Technologies used:

DICOM Viewer, PACS, mathematical algorithm for coloring gray areas on DICOM images

Task:

Establishing links between the highlighted colored area and the real area of pathology. The main task is to investigate the relationship between the selected segment of grayscale in the image and the pathology itself in order to achieve fast high-precision marking independent of the human factor in terms of the accuracy of the selection and detection of the pathology segment

Solution:

Web application (DICOM-image markup, DICOM-attribute markup (metadata))

DICOM image processing (Based on grayscale (256 - 65536 shades))

Result:

Frosted glass area: 180.770 ml Consolidation area: 11.472 ml Pleural effusion area: 53.084 ml Volume of the affected lung area to the total lung volume: 66%

4. AI recommendation systems for the diagnosis of diseases (malignant formations)

Technologies used:

Frameworks: Tensorflow, PyTorch Segmentation NN: U Net, W Net, Mask R-CNN, MeshNet, CNN+CRF Language: Python 3

Result:

The probability of ZNO throughout the image is 92.03%

Task:

Develop, train and combine into a common ensemble of algorithms for processing medical data (including DICOM images, DICOM tags, patient metadata (including formalized protocols)

Achieving classification accuracy of at least 95% per class (for the task of classifying images/studies)

Achieving the accuracy of the pathology segment selection in the image is higher than 90% (for the segmentation task)

AI module for the diagnosis of diseases:
An INS-based model for metadata processing (Based on the analysis of DICOM tags, Based on the analysis of patient metadata (formalized protocols)) An INS-based model for processing DICOM images (Segmentation of pathologies in the image, medical research, Diagnostics (classification of the image))

Found 1 hearth with a volume of: - 7.2 ml

Date of analysis - 2021-02-04

Ready to start something great?