Assisted Intelligence in Medicine
using Interactive Graphs
AIMIG.org
ES, version 2026-08-14
- Problem
- Healthcare is far from optimal as well in developing
regions as well in developed ones for many reasons including
lack of appropriate access to knowledge, delays for access
to doctors, budget constraints, poor availability in rural
areas. No individual doctor can any more master all medical
knowledge domains, including "rare disease".
- Mission
- Make medical knowledge better available for
decisions in a practical way for patient care.
- Moreover useful for education, research, care quality and
cost limitation.
- Approaches
- "AI" as explainable "Assisted Intelligence" rather than
"Artificial Intelligence".
- Decision support based on a Medical Knowledge Model,
maintained by a team of experts.
- Complex medical knowledge can be represented by mean of
graphs. Concepts as "nodes" and relationships between
concepts as "edges" where any node can have a relationship
with any other nodes in a space of millions of concepts and
where relationships may be qualified. Decisions can take
account of many factors. In fact graph knowledge models
allow a kind of computer simulation of the logic of human
mind with many neurons linked by synapses.
- To be seen as assistance from a team of consultants, where
one consultant is a computer system representing the
know-how of remote experts.
- A task oriented approach, identifying problems and seeking
solutions.
- Medical
Knowledge as graph
- Better access to already existing medical knowledge.
Conversion of current medical knowledge into graphs. All the
information useful for decision support.
- A synthesis from different sources as medical, courses,
textbooks, ontologies, literature, and above all medical
experts from specialized scientific communities.
- A large amount of biomedical knowledge is in
principle already available but the question is now how to
use this knowledge in a more efficient way.
- Focus on the relations between symptoms, problems and to
be recommended actions.
- Maintenance of agreed medical know-how, based on an
international community of experts.
- Patient
record as graph
- Patient information also structured as a graph with links
to related knowledge. For example easy navigation from every
"health issue" to symptoms, complication, actions,
patient-doctor encounter, ...
- Decision
support engine
- Given information from both the patient and from
medical knowledge, try to provide recommendations. Using
"graph navigation", "vectors", "agents", etc...
- A " Decision Engine " with focus on rules about how to
process knowledge graphs and patient graphs.
- Qualified recommendations with probability, degree of
certainty, priority, potential risks, expected benefits and
side effects of possible therapies, ...
- What most matter here is the medical logic with
explanations.
- Intended users:
- Typical Use Case:
- A patient arrive in emergency room with a problem, for
example shortness of breath, what are the likelihoods of
possible issues and what are the relative priorities of
what should be done next?
Step by step priorities may be of giving oxygen ? of which
most relevant questions ? which physical examination ? ask
thorax xRay images ? ask an ECG ? ask lab tests ? begin a
treatment ? decision about admission ?
- After every new information, re-evaluate the new
situation and adapt the visual graph of likelihoods and
priorities.
- Telemedicine in situations where there is limited local
qualifications and/or not enough doctor time per patient.
- Contribution to medical education and training, in front
of interactive decision strategies.
- Preliminary sorting of patients on a waiting.
- Reduction of doctor time spent on basic information
gathering.
- Many specialized health organizations are working on
knowledge bases in traditional and incompatible ways.
The challenges are integration and accessibility in a
format suited for decision support.
- Quality checks which may provide warnings, even in most
advanced countries.
- Human
graph interface
- The logic of the project is essentially based on graphs.
Why not an interface based on graphs rather than only on
texts ? Both humans and machine can understand and draw
graphs.
- Education
- Involvement of teachers and training of students playing
with graphs in order to discuss differential diagnosis and
the potential benefits of considered next actions.
- Research
- At a later stage, analysis of large populations of patient
records can improve the knowledge base. Fine-tuning of the
weights of relations between concepts. Evaluation of the
results of treatments.Discovery of unsuspected patterns
using graphs software tools.
- Experimental prototype
- Call for a community of physicians, data scientists and
software developers.
- Medical:
- Knowledge:
- Conversion of medical Knowledge into graphs, at
least in some medical domains,. Not only usual graphs
of coagulation factors, but knowledge about dyspnea ?
chest pain ? abdominal pain ? meaning of lab tests ?
etc .... . Focus on the properties of relations
between concepts.
- Interactive decision support:
- Conversions of patient record into graphs. Make
links between observed symptoms of the patient and
theoretical knowledge about these observations.
- Experimental decision support. Initially limited to
decisions about which additional information to seek
next.
- Evaluation about how the know-how of top medical
experts could be made available and useful to beginners
in remote regions.
- Technical:
- Make a better "Interactive Graph Editor" in order to
facilitate the man-machine communication. Already
possible but laborious.
- Analysis of the logic of medical reasoning and
development of decision support software tools.
- Organization::
- Seeking grants for this international Open Source and
Open Data not-for-profit initiative, with support from
scientific communities, universities, ...
- However "support services" remain usual business for
every healthcare organization, including installations,
training and maintenance.
- Contact: etienne@saliez.be, ...