There are three major areas of medical informatics related to the cardiovascular system:
- Automatic analysis of ECGs
- Imaging of the heart and blood vessels
- General cardiovascular diagnosis/decision aids
Why
is automatic ECG analysis difficult?
- It must have a high sampling rate (especially to get R and S waves )
- There is great variation within the population
- Usually you need to know what a person's "normal" is before you can say if
it is abnormal
- There are some ECGs waveforms that are usually easy to diagnose (a-fib,
v-fib, etc)
- Calibration
- Voltages of the waves in the normal ECG depend on the manner in which the
electrodes are applied to the surface of the body
- There is variation in where the electrodes are placed.
- If one electrode is placed directly over the heart the voltage of the QRS
might be 3mV!
- If electordes are placed on 2 arms or one arm and one leg, get appropriate
1mV.
Arrhythmia
Monitoring of ECGs
- ECG arrhythmia-monitoring systems have been accepted quickly
- one of the most sophisticated and difficult of the bedside monitoring
tasks
- Old system - people watch
- expensive, unreliable, tedious, and stressful to observers!
- Early system - time-shared central computers
- "New" systems
- built in arrhythmia monitoring systems
- 16-bit architecture
- waveform templates
- real-time cross-correlation techniques to classify rhythm abnormalities
- Computer Algorithms for processing ECG rhythms
- Larsen and Jenkins, 1987
- extract features such as amplitude and duration of the QRS complex
- performs feature extraction by serching the sampled data from start 2 end
to locate the P wave and the QRS complex, and the T wave
- slope-detection often used to detect quick upswing of qrs
- thresholds value of slope -- sometimes user can input thresh
- p and t harder to detect because of gentle slopes
- difficulty in detecting p = significant problem
- noise = other problem
Commercial
ECG Monitors
- Marquette 7700 series bedside monitor
- samples the 4 leads simultaneously
- before performing the qrs detection, looks for noise, then waits 2.5 secs
- when monitoring begins there is a learning process to determine which qrs
waveform shape is seen most frequently
- 16 beats to determine R-R intervals and calculate the average
- compares with templates to determine abnormalities
- 4 levels of alarms depending on severity of arrhythmia
- Others: HP, GE, Patient Data Management Systems, etc.
- emphasize acquisition and processing of physiologic data noton integration
w/other depts
See also:
- Imaging the heart and the chambers within it is incredibly difficult
because the heart is always moving!
- Imaging vessels is also difficult because they are hard to see, and it is
hard to tell what is connected to what when you do see them.
Angiography
- imaging of the blood vessels performed by the injection of opacifying
agents into the bloodstream
- experiments conducted as early as 1923
- Conventional angiography
- inject contrast in peripheral vein
- but --> lung and heart and is too weak when it gets there
- put catheter in artery and thread it up to the bifurcation
- CT and MR Angiography
- Digital Subtraction Angiography (DSA)
- DSA increases visualization of low concentration of contrast
- record mask of vessels before contrast isinjected
- inject
- subtract mask image from current images (only difference should be where
contrast accumulated)
- in practice, still must do arterial inject, but less contrast
- good for head because you can subtract out the skull
- Angiography also gives you information on flow
- The actual output is a velocity vector in terms of velocity in terms of
(x, y, z).
- Usually just take the magnitude of this velocity to show where there is
movement (i.e. blood)
- Some applications, however, try to use this velocity information:
- Maria Tovar - Stanford MIS student
- - 3d visualization of vasculature using MRA
- - use velocity information to determine connectivity of vessels as well as
to determine vessel boundaries etc.
- Sandy Napel - Stanford Radiology
- - work to actually visualize flow
- - that way you can tell where an artery is blocked
- - also helps you to determine connectivity of vessels
See also:
III. General cardiovascular diagnosis/decision aids
- Homer Warner in 1964 at LDS
- Approach:
- Uses a big matrix to store all of the a priori & conditional
probabilities
- Uses Bayes' theorem to calculate the probabily of disease given symptoms
- Outputs a differential diagnosis (probability for each disease)
- Assumptions:
- Conditional independence: symptoms are independent of one another within
given disease
- Mutual exclusivity: each patient could only have one disease
- If patient is found to have more than one disease then the new combination
must be considered as a new disease entity
- Results:
- Test data was 53 symptoms & 35 diseases
- They concluded that the diseases could be diagnosed with an accuracy equal
to that of an experienced specialist in the field
- Would most likely get better with further refinements to the disease
matrix
- Problems:
- Limited to combinations for which probabilities were available
- Assumptions not always valid
- Difficult to obtain accurate probabilites
- Prior probabilities will be different in different geographic
locations
- Gorry (MIT) & Barnett (MGH), 1968
- Approach:
- Similar to Warner's approach but uses sequential diagnosis
- Constructs "current view" of problem based on what's known so far
- Weighs expected value of tests against costs (test price, risks of tests,
pain, consequence of misdiagnosis)
- Determines when to stop
- Uses Warner's data
- Problems:
- Difficult to estimate costs
- Same problems as Warner's system
- Baxt (UCSD), 1991
- Inputs:
- History: age, sex, location of pain, intensity of pain, nausea, vomitting,
diaphoresis, syncope, shortness of breath, palpitations, response to
nitroglycerin
- History findings: MI, angina, diabetes, hypertension
- Exam findings: rales, jugular venous distension
- ECG findings(!!!): 2mm ST-segment elevation, 1mm ST-segment, elevation,
ST-segment depression, T-wave inversion, significant ischemic change)
- Results:
- Prospective study of patients in the ER with chest pain
- Compared with physicians
- Found increased sensitivity and specificity
Digitalis Therapy Adviser
- Gorry, Silverman, Pauker, 1978
- Based on lots of models:
- Causal models of heart rhythm disturbances and principles of
antiarrhythmiatherapy
- Pharmokinetic model of digitalis
- Patient-specific model incorporating qualitative information
- Temporal Reasoning and Planning
- Uses the results of previous treatments to alter the model of the patient
- Ex. If predicted level of digitalis is higher than the measured stores,
then the system would adjust the "oral absorption" paramter down.
XPLAIN
- Swartout, MIT, 1981
- explanation for the Digitalis Therapy Adviser
- based explanation on abstract principles: domain model
See also:
- ECG (or sometimes EKG) is short for electrocardiogram.
- Using electrodes placed on the skin on opposite sides of the heart, an ECG
measures the electrical potentials of the heart.
- A normal ECG is made up of:
- P wave- caused by the spread of depolarization through the atria (followed
by the atrial contraction)
- QRS waves - result of the depolarization of the ventricles (prior to
ventricular systole)
- ventricular T wave - stage of repolarization of the ventricles -->
muscle begins to relax (slightly prior to end of ventricular
contraction)
- heart or its associated blood vessels are malformed during fetal life
- some are hereditary
- one common cause is a virus infection in the mother during the first
trimester of pregnancy when the heart is being formed (particularly if mom
contracts German measels at this time!)
- Ex. anomalous origin of coronary artery, ruptured sinus valsalva, total
anomalous pulmonary venous connection, atrial septal defect, pulmonary
stenosis, etc.
- Immediately after an acute coronary occlusion, blood flow ceases in the
coronary vessels beyone the occlusion except for small amounts of collateral
flow from surrounding vessels
- the area of muscle has zero or little flow --> cannot sustain
cardiac muscle function (infarcted)
- NB: this diagnosis is very important...tend to hospitalize patients with
chest pain more often than necessary just to be on the safe side (i.e. it is
bad news to send someone with a heart attack home!) --> big cost
- class of drugs used to treat heart conditions
- ex. digoxin and digitoxin
DISCLAIMER: The information on these pages is not guaranteed to be correct nor should it be used to make any health care decisions.
Created by Wanda Pratt
Much of the original text by Rhea Tombropoulos
Inspired and assisted by Russ Altman