Portfolio slaideris sākumlapa ENG

AI-DRIVEN ISCHEMIC STROKE DIAGNOSTIC SYSTEM

Project background:

The rising number of radiological exams highlights the pressing need for faster and more accurate diagnoses.

Biggest challenges: Complex data description and computer vision training due to the wide variation in data – different features may be present for the same diagnosis.

Solution:

The solution is a prototype tool that uses computer vision algorithms and a set of neural network models to perform a fully automated analysis of a brain CT scan. System can identify possible brain pathologies and determine if a patient is experiencing an acute stroke. It is able to accurately identify, classify and localise acute stroke, other types of abnormalities and norm.

The solution consists of several components that perform following processes: automated reception of data from the hospital CT scanner, anonymization of the data, and data transmission to an AI service for processing. The AI processing service performs automated processing of complete examination, analysing each layer of the examination, describing results and sending them to the hospital’s internal application, which is used to view and analyse patient data.

AI component of the solution consists of computer vision algorithms and neural network models that are specifically designed and trained for the solution. This was the result of an extensive study that led to the development of a complex solution.

Solution also includes a common data description methodology and a common data description tool for maximum accuracy and quality to ensure the highest possible data purity.

The solution has been integrated into the PSCUH (Pauls Stradiņš Clinical University Hospital)system since March 2024 and it is being tested and used as a tool to support daily work in diagnostics.

Key benefits:

• The Acute stroke diagnostic system is already able to work with 93% accuracy, which is far more accurate than experienced diagnostician;
• Time for diagnosis is shortened, which can be crucial in cases of ischemic stroke;
• Decreased workload for medical personell in the reception department.

AUTOMATED FOREST AUDITING SYSTEM WITH A PORTABLE MEASUREMENT TOOL

Project background:

Forest surveying and auditing are done manually – person uses a tape measure to determine the parameters of each tree and then calculates the forest’s value. Client wants to automate forest surveying and auditing process.

Biggest challenges: Difficulties in accuracy and speed of data acquisition: tree canopy and forest density make it difficult to collect data quickly, and smaller trees move in the wind, which makes data acquisition more difficult.

Solution:

An automated forest measurement and auditing system, as well as a portable, automatic forest measurement tool have been developed as part of the project. The solution is designed as a backpack equipped with special sensors. By walking through the area along a pre-established route, all the necessary data is obtained to calculate forest value. The portable tool performs data collection, automatic data processing, and it also retrieves results after data collection. The system can identify trees, classify tree species, and determine their volume, replacing manual measurement methods where a person measures each tree and performs approximate statistics-based volume calculations.

Key benefits:

• Automated data collection and processing – data collection time reduced 5-10 times;
• Reduced time needed for data extraction;
• Improved data accuracy – increased to over 95% (from previous 85-90%);
• Optimised workforce and efficiency.

ROAD ASSET MANAGEMENT SYSTEM

Project background:

The client has several databases containing different types of data and lacks an up-to-date inventory and management system mapping the main national roads and their adjacent assets.

Biggest challenges: To configure  a mobile data acquisition vehicle with sensors, connect them to navigation equipment and synchronise them with business management system. The project requires highly accurate data, yet navigation system is subject to biases and variability in the data.

Solution:

Developed an AI system that can automatically perform a full analysis of the road and its assets by analyzing data collected on the road.

A specialized vehicle equipped with sensors, video data streams, and navigation data is used to combine these elements for automated analysis, without human intervention.

A mobile device – car – is specially equipped with necessary sensors to collect the data. Systemreceives video streams from different sensors as input, combines them with navigation data and performs automated analysis of the road and adjacent objects, by locating and classifying certain clustered objects in the image. The precise location of each object can be extracted. The solution can perform fully automated data analysis and information extraction using an AI component without requiring manual intervention.

System consists of several modules – System App, Data Collection App, Data Upload App, AI Module or Data Processing Module. The AI module performs full analysis of road and road asset data, carrying out road asset information segmentation, tracking, object classification, precise object location determination, orthophoto and 360-degree view generation.

To achieve this, solution incorporates a set of complex computer vision algorithms, as well as various machine learning solutions and models, all of which are uniquely designed to address this problem. Several custom developed and trained neural network models are used for the solution, utilizing data sets and customized neural network structures.

Key benefits:

• Improved accounting, management, and related business process administration system and data reliability;
• Simplified system management;
• Ensuring data accuracy and information availability.