Full TitleComputer Assisted Thoracic Assessment using POCUS
Lung ultrasound is a relatively recent medical technique, enabling physicians to quickly monitor and diagnose cases where alternative tests are impractical and cumbersome, and enabling simple and fast decisions such as patients triage to hospital emergency rooms or intensive-care units. Point-of-Care Ultrasound (POCUS) is particularly interesting in the COVID-19 context, given its portability and bedside use, enabling doctors to identify the viral pneumonia associated with the most serious cases of this virus, by inspecting the thickening of the pleural line, and the presence and characteristic patterns of B lines.
However, widespread use of this technique is limited by the operational complexity and inter and intra-observer variability of a POCUS exam. Correct probe positioning and image assessment are two interlinked complex tasks, requiring expert knowledge and training for effective use.
This creates an opportunity for exploring the potential of Artificial Intelligence (AI), more specifically, computer vision based on deep learning architectures, as an enabler of POCUS for massive use. What if we could automatically guide an inexperienced user in positioning a POCUS probe for COVID-19 screening and assessment, and give him a computer assisted diagnostic (CAD) suggestion on the spot?
The THOR project (Computer Assisted Thoracic Assessment using POCUS) is a multi-disciplinary collaboration between INESC TEC (computer vision), Hospital Garcia da Horta (medical), FC.ID (computer vision) and Nevarotech (interoperability, prototyping, technology transfer), with the main goal of creating a proof-of-concept prototype of a computer vision CAD system for COVID-19 screening and assessment using POCUS.
Accomplishing it, will require addressing of the following objectives:
1. Create a dataset of annotated POCUS exams and relevant associated data, which can support the research of the novel algorithms envisioned for the project.
2. Research and develop novel computer vision algorithms for the creation of the assisted probe positioning functionality.
3. Research and develop novel computer vision algorithms for the creation of the screening and assessment functionality.
4. Engineer an early stage prototype of a CAD system, which integrates the developed algorithms, and has enough maturity to be installed, tested and demonstrated in a real environment.
Achieving these objectives requires a solid foundation on this topic, an adequate strategy and an experienced multi-disciplinary team.