Development and commercialization of a semi-supervised learning AI for robust diagnosis in real world settings.

Projektinformationen

MrDoc

ID Finanzhilfevereinbarung: 876145

DOI

10.3030/876145

Projekt abgeschlossen

EK-Unterschriftsdatum 26 Juli 2019

Startdatum 1 August 2019

Enddatum 31 Januar 2020

Finanziert unter

• INDUSTRIAL LEADERSHIP - Innovation In SMEs

Gesamtkosten

€ 71 429,00

EU-Beitrag

€ 50 000,00

Koordiniert durch

MR DOC SRL
Italy

Periodic Reporting for period 1 - MrDoc (Development and commercialization of a semi-supervised learning AI for robust diagnosis in real world settings.)

Berichtszeitraum: 2019-08-01 bis 2020-01-31

Zusammenfassung vom Kontext und den Gesamtzielen des Projekts

Non-communicable diseases are the leading cause of death in the world, and doctors struggle to properly address the problem, due to the aging population trend and the still insufficient number of physicians per 100,000 people.
In Europe, the economic burden of NCDs is growing for both patients and healthcare providers.
This situation highlights the need for technological solutions supporting mass prevention.
AI-based diagnostic software proved able to contribute to early detection of diseases or health risk situations, but they come along with many significant limitations, such as the need to train on huge amounts of labeled data and difficulties in managing inputs that are noisy, incomplete or simply different from the original dataset.
We designed a solution overcoming the said limitations, thanks to a semi-supervised learning AI, able to analyze and interpret the medical dataset through a process that mimics human creative imagination. The final goal is to detect and diagnose some NCDs (like arrhythmias, hypertension) and some biometric parameters (blood pressure, Heart rate variability (HRV), hemoglobin, blood glucose) from “dirty” signals, generated by consumer electronics devices such as smartphones and closed-circuit cameras, and to detect pathologies and measure relevant volumes and diagnostic signs on clinical images generated by medical devices.

Arbeit, die ab Beginn des Projekts bis zum Ende des durch den Bericht erfassten Berichtszeitraums geleistet wurde, und die wichtigsten bis dahin erzielten Ergebnisse

During the action, we aimed to validate the solution in a feasibility study made up of a marketing strategy, a report on technical development requirements and risks and an assessment of the IP and regulatory landscape. We engaged also end-users and partners to gather precious feedbacks and evolve the design towards a commercially ready MrDoc product.
The result of the action was a Phase 2 Project Plan, along with an upgraded business model and updated financial projections.
Among the most significant results obtained, the most remarkable are: i) the growth of the diagnostic AI platform into an Active Learning system, able to interact with medical doctors in an iterative process, and ii) the improvements in the “cross normalization” layer, able to normalize data coming from different devices and sources, to make it easier to manage for the diagnostic AI.
Along with that, significant step forwards were performed in signing new partnerships, achieving traction, gathering further clinical results and developing new sources of data acquisition, such as “video-selfie” PPG signal extraction from a smartphone camera.

Fortschritte, die über den aktuellen Stand der Technik hinausgehen und voraussichtliche potenzielle Auswirkungen (einschließlich der bis dato erzielten sozioökonomischen Auswirkungen und weiter gefassten gesellschaftlichen Auswirkungen des Projekts)

The following technical developments have been made:
1. from a scientific standpoint, we validated on a reasonable number of subjects the possibility to:
- Extract an accurate PPG from the contactless video stream of patients face instead of that from a finger touching a camera;
- Detect not only arrhythmia but hypertension and anemia from the said PPG, achieving an accuracy superior to the EU standards for medical devices.
2. from a technical standpoint, we realized:
- An API service, able to integrate the platform with external UX interfaces, such as apps, websites or desktop applications
- A new, more user-friendly UX integrating an advanced face recognition system detecting in real-time the presence and correct positioning of a face in the camera
- A back-office website allowing doctors to interact with patients regarding measurement results, with an integrated messaging system
- An Active Learning platform allowing doctors and AI to interact collaboratively and iteratively in the process of annotating and labeling a medical dataset, restricting human intervention on the most representative and significant samples only, and letting the AI learn on its own on the remaining samples
- A new tier featuring feature extraction from medical images and their segmentation
- Training of the AI on new health indicators, problems and noise types.
These significant results will contribute significantly to improve diagnostics accuracy and time and cost associated with it, achieving an overall unprecedented goal for patients and healthcare providers.

MrDoc Semi-supervised learning in medical imaging