Periodic Reporting for period 1 - Doctomatic (Development of advanced AI algorithms for remote patient monitoring)
Période du rapport: 2023-06-01 au 2024-01-31
Current standards of care of remote patient monitoring solutions require painful integrations with medical devices that make the service tough to handle and inoperable at some points. We still find patients taking their measurements on paper notes that usually get lost by the time they have their next medical appointment. Readily available remote patient monitoring softwares are usually focused on one healthcare condition only and require end users to purchase new medical devices with the software preinstalled for data transmission making it complex for healthcare providers and costly to both national healthcare systems and patients.
Doctomatic provides a unique solution using visual IA for multiple healthcare conditions that allows to transform images of the results of measurements of medical devices into data (heart rate monitors, scales, thermometers, glucometers, pulsioximeters).
Doctomatic has been used in one of Barcelona’s public main hospitals (Hospital del Mar, currently in pause waiting for the public tender to be published), private medical practices and Clariane group owned “Seniors Residencias” (elderly care homes) in Spain. One of our use cases is a patient with cardiac insufficiency who requires frequent monitoring of both their heart rate and weight. The doctor prescribes the use of Doctomatic either as a stand-alone app or embedded within the HCP providers’ own app to capture their measurements on a regular basis. Patients would get a reminder, take their
measurements and send the data. If the data is non-existent or out of range, the doctor/medical centre is alerted and can act accordingly preventing the worsening of that patient’s condition.
Our TRL is 9, however our AI technology needs more development in order to read more devices and with more accuracy. Doctomatic currently reads the images of: glucometers, heart rate monitor, thermometre, digital scales, pulse oximeters and Alivecor Kardia ECG reports. Our challenge is to be able to detect more devices that have never been connected and allow them an extended digital life in the preventative healthcare market such as the peak flow (asthma & COPD), urine reactive strips, etc.
Doctomatic provides a unique solution using visual IA for multiple healthcare conditions that allows to transform images of the results of measurements of medical devices into data (heart rate monitors, scales, thermometers, glucometers, pulsioximeters).
Doctomatic has been used in one of Barcelona’s public main hospitals (Hospital del Mar, currently in pause waiting for the public tender to be published), private medical practices and Clariane group owned “Seniors Residencias” (elderly care homes) in Spain. One of our use cases is a patient with cardiac insufficiency who requires frequent monitoring of both their heart rate and weight. The doctor prescribes the use of Doctomatic either as a stand-alone app or embedded within the HCP providers’ own app to capture their measurements on a regular basis. Patients would get a reminder, take their
measurements and send the data. If the data is non-existent or out of range, the doctor/medical centre is alerted and can act accordingly preventing the worsening of that patient’s condition.
Our TRL is 9, however our AI technology needs more development in order to read more devices and with more accuracy. Doctomatic currently reads the images of: glucometers, heart rate monitor, thermometre, digital scales, pulse oximeters and Alivecor Kardia ECG reports. Our challenge is to be able to detect more devices that have never been connected and allow them an extended digital life in the preventative healthcare market such as the peak flow (asthma & COPD), urine reactive strips, etc.
Different activities were carried out during the project to improve Doctomatic’s accuracy.
Selection of AI architectures.
Compare different existing AI architectures evaluating their performance to choose the most suitable one for Doctomatic.
Different metrics were obtained and a specific architecture was chosen.
Automatic creation of datasets
An automatic dataset creation process has been established. These datasets are then used to train the AI and optimise its performance.
To create the datasets a specific type of algorithms have been used. These are called genetic algorithms, a type of optimization and search algorithms inspired by the principles of natural evolution and genetics.
AI training agent generator
A web service has been developed in Python using the chosen AI architecture and the created datasets. This service is the core of Doctomatic, as it is able to recognize the digits in the images and translate them into numbers.
Creation of Cognimatic - an extra layer of abstraction around the AI architecture
An extra layer of abstraction has been built around the chosen AI architecture. This allows our system not only to recognise the digits, done thanks to the AI architecture, but also to identify their position and turn them into meaningful values or figures.
Evaluation agent software of trained AI solutions and post training evaluation
A specific software was created to compare AI architectures with varying parameter configurations using different datasets. This allows us to determine which specific models improve the object identification performance task. The metric compared is the accuracy of the different models, and the result is represented in terms of percentage of correctly annotated images. This information can also be given by type of medical device. Specific architectures are then chosen based on the results obtained.
Different versions of the Doctomatic system as a whole can also be compared with this software.
Selection of AI architectures.
Compare different existing AI architectures evaluating their performance to choose the most suitable one for Doctomatic.
Different metrics were obtained and a specific architecture was chosen.
Automatic creation of datasets
An automatic dataset creation process has been established. These datasets are then used to train the AI and optimise its performance.
To create the datasets a specific type of algorithms have been used. These are called genetic algorithms, a type of optimization and search algorithms inspired by the principles of natural evolution and genetics.
AI training agent generator
A web service has been developed in Python using the chosen AI architecture and the created datasets. This service is the core of Doctomatic, as it is able to recognize the digits in the images and translate them into numbers.
Creation of Cognimatic - an extra layer of abstraction around the AI architecture
An extra layer of abstraction has been built around the chosen AI architecture. This allows our system not only to recognise the digits, done thanks to the AI architecture, but also to identify their position and turn them into meaningful values or figures.
Evaluation agent software of trained AI solutions and post training evaluation
A specific software was created to compare AI architectures with varying parameter configurations using different datasets. This allows us to determine which specific models improve the object identification performance task. The metric compared is the accuracy of the different models, and the result is represented in terms of percentage of correctly annotated images. This information can also be given by type of medical device. Specific architectures are then chosen based on the results obtained.
Different versions of the Doctomatic system as a whole can also be compared with this software.
Having selected an appropriate AI architecture, and being able to train it with created datasets, the system has achieved an accuracy of 95% in correctly identifying the data in different types of medical devices.
This enhancement directly translates to increased user satisfaction and improved usability. This, in turn, drives widespread adoption, positioning Doctomatic to seamlessly scale and serve a significantly larger user base.
The key needs to ensure further uptake and success would be to ensure access to other markets such as France, the U.K or the U.S. by defining a proper commercialisation strategy. Moreover, to deploy pilot studies Doctomatic intends to comply with the Guideline on Computerised Systems and Electronic Data in Clinical Trials proposed by the European Medicines Agency. A guide whose objective is to ensure that data generated in clinical trials is accurate, reliable, and of quality, regardless of whether it is collected electronically or on paper.
Regarding entrance in the US, Doctomatic would also like to comply with the FDA 21CFR Part 11 which establishes the requirements for the security and integrity of electronic records and electronic signatures used in computer systems in industries regulated by the FDA.
This enhancement directly translates to increased user satisfaction and improved usability. This, in turn, drives widespread adoption, positioning Doctomatic to seamlessly scale and serve a significantly larger user base.
The key needs to ensure further uptake and success would be to ensure access to other markets such as France, the U.K or the U.S. by defining a proper commercialisation strategy. Moreover, to deploy pilot studies Doctomatic intends to comply with the Guideline on Computerised Systems and Electronic Data in Clinical Trials proposed by the European Medicines Agency. A guide whose objective is to ensure that data generated in clinical trials is accurate, reliable, and of quality, regardless of whether it is collected electronically or on paper.
Regarding entrance in the US, Doctomatic would also like to comply with the FDA 21CFR Part 11 which establishes the requirements for the security and integrity of electronic records and electronic signatures used in computer systems in industries regulated by the FDA.