2. 02 Healthcare data in Moscow – current state
development of citywide Integrated Medical Information Analytical System started…
6 years ago
Now IMIAS includes:
Infrastructure Services
For citizens
Online appointments
For physicians
Electronic medical
record (EMR) —
collecting of patient electronically-
stored health Information
Control room —
everyday gathering
and processing of health data
e-Desk —
remote patient flow management
e-Prescribing —
automated procurement of medicines
Sick notes —
automated system
of medical certificates recording
Laboratory service —
providing a set of clinical diagnostic tools
700+
health facilities
integrated to the system
23 000+
medical workstations
21 000+
healthcare professionals
9 mln+
unique patients
3. 03 Healthcare data in Moscow – current state
All these data gives unique possibilities for processes’
improvements and decision supporting
Areas of joint activities by IT and Healthcare departments of Moscow:
Data science in healthcare
Data-driven operations and processes optimization
Demand prediction
models for clinics
Computer vision medical
recognition systems
Citywide epidemics
prediction models
Physicians work schemes
analysis and optimization
Balancing of human
flows between clinics
4. 04 Applied data science in healthcare of Moscow
Why do we actually need recognition systems?
97%
target success rate
of the recognition of the best
doctors in the world
It is impossible to bring
all doctors to that level
It is possible to train neural
network to reach such level
#1
in the world of all malignant
formations revealed on computer
tomography images
Main directions of recognition systems
development by Moscow government
#1
disease-related death reason
in Russia, symptoms could be
revealed on computer
Neoplasms
in lungs
Cardiovascular
diseases
5. 05 Neoplasms in lungs recognition
80%
Of all lungs cancer cases
detected on 3rd-4th stage
(complex and hard to treat)
Current state of neoplasms in lungs treatment in Moscow
40%
Patients survive
throughout next
12 month
10%
Patients
could be treated
*100 patients
Out of which 50% Out of which 20%
Project key steps and current state
stage1
SEGMENTATION
Selected an organ
in an image
stage2
RECOGNITION
Determination of anomalies
using neutral network
stage3
CLASSIFICATION
Determining
which of the anomalies
are malignant and why
86.5%
of all tumors
were recognized
in the test sample*
6. 06 Neoplasms in lungs recognition - algorithm
DICOM is a series of images We collect a 3D cube
with all the organs
Cut out only the organ needed
for analysis without interference
All the output, the image is restored
and based on the sum of the parameters
assigned by the grid, each sender is determined
the probability of a neoplasm in each region
On each layer of the network,
the imageis transformed and
it’s parameters are selected
Split the images into many
small pieces. Feed pieces of
3D organs into the network
7. 07 Cardiovascular diseases recognition
20%
Fully recover
35%
Fully recover
60%
Doesn’t come back
to productive life
20%
Patients die
Every year 30 000 people endure
ischemic stroke treatment in Moscow
Current major problems
with cardiovascular diseases
In same period 100 000 people endure Miocardial infarction
No systematic work according
strokes’ and infractions’ symp-
toms detection is conducted
ECG data is not stored
in IMIAS for further analysis
No systematic monitoring of
risk-group conditions is conducted15%
Patients die
50%
Doesn’t come back
to productive life
8. 08 Cardiovascular diseases recognition - algorithm
ECG signal
All ECGs regardless their
target are stored in IMIAS
and being automatically
analyzed for pathologies
Neural network
Allows to gather healthcare
knowledge and sustainably
improve quality of analysis
Diseases detection
Improvement of treatment and
prevention quality. Strong pro-
spectives for further evaluation
of wider list of pathologies
Target monitoring process
Recognition algorithm
An additional layer of architecture: converting the ECG signal into a spec-
trum and applying image recognition techniques to the spectrum picture
(similar to voice recognition technology). An effective solution of the CAC
group. Previously, such a method was not published anywhere