Overview
Sirma developed Diabetes: M to simplify blood sugar tracking for diabetic patients by enabling users to capture a photo of their glucometer screen instead of manually entering data. This innovation reduces barriers for patients with limited technological access or skills, eliminating the need for costly sensor devices or manual logging.
The Challenge
Managing diabetes requires accurate and consistent blood sugar tracking, which is often hindered by challenges such as patients’ unfamiliarity with digital health apps, the inconvenience of manual data entry, and the high cost of sensor technologies. These issues lead to poor adherence and less reliable health monitoring, impacting diabetic patient outcomes. The challenge was to create an intuitive, affordable solution that lowers entry barriers while maintaining data accuracy and usability.
The Project Scope
Sirma has added a new its user-friendly feature to its mobile application, which was capable to:
- Capture and digitize blood sugar readings from photos of diverse glucometer devices;
- Bypass manual input, enabling inclusive use by patients with low digital literacy or limited device access;
- Process images quickly and accurately to extract glucose data;
- Provide seamless integration with existing diabetes management platforms and healthcare workflows;
- Ensure privacy and security standards are met.
The Solution
Sirma implemented an AI-driven image recognition system that leverages computer vision algorithms to detect and extract blood sugar readings from glucometer screen captures. Key solution components include:
- Optical Character Recognition (OCR) optimized for device screens under variable lighting and angles
- Machine learning models trained to recognize digits and display formats for a variety of glucometer brands
- A mobile-friendly user interface optimized for simple photo capture and instant feedback
- Backend infrastructure supporting secure, real-time data processing and integration with health records
- Accessibility-focused design ensuring usability across diverse patient populations
Results
- Simplified data entry process, reducing patient burden and increasing adherence to blood sugar monitoring.
- Broadened accessibility for patients without expensive sensor devices or high digital proficiency.
- Enhanced data accuracy by automating glucometer readings capture, reducing human error from manual logs.
- Improved patient engagement and more reliable blood sugar tracking contributing to better diabetes management outcomes.
Technologies
- Computer Vision and OCR: Tailored optical character recognition algorithms capture and extract data specifically from medical device screens, including various glucometer displays;
- Machine Learning for Device and Display Recognition: Models are trained on extensive datasets of images representing different glucometer models, screen sizes, display styles, and environmental factors (lighting, angles) to identify the device and customise processing accurately;
- Image Preprocessing: Techniques such as image enhancement, normalization, and correction of distortions (glare, shadows, skew due to camera angle) prepare images for optimal OCR performance;
- Digit Localization and Parsing: Algorithms pinpoint the exact locations of glucose readings on device displays, detecting characteristic screen segments or pixel patterns to extract numeric values reliably;
- Mobile Application Framework: The app is optimized for accessibility and ease of use, enabling users to snap clear photos of any glucometer with a simple, intuitive workflow;
- Continuous AI Model Retraining: The system incorporates new images and user feedback to continuously improve recognition accuracy, support new glucometer models, and adapt to evolving display designs;
- Robust Error Handling: Confidence scoring mechanisms flag uncertain or low-confidence readings for manual review or prompt users for additional image captures, minimizing errors.
Sirma’s Partnership with the client
Sirma utilized its expertise in AI and mobile technology to create effective healthcare solutions for diabetes management. The new feature in Diabetes: M processes images from various glucometer models using advanced computer vision and machine learning tailored for different devices.
