Algorithmic Approach to Chest Disease Diagnosis

An algorithmic approach to Chest Disease involves following a decision tree. For example, the answer given to whether the chest disease is "Focal or Diffuse?" will lead to two divergent pathways of questions to help narrow down the possibilities. This approach makes sense because different diseases can present differently.

Let's look at the example published by two Canadian physicians, Drs. Ginzburg and Rahman, (used with permission).

Algorithm Chart Differential Diagnosis of Parenchymal Lung Disease

This is a single sheet with numerous pathways. It is a large file but you may wish to view it anyway.

Let's take an example.

Suppose you are looking at a chest x-ray.

The algorithm first asks you do decide if the x-ray is NORMAL or ABNORMAL.
- Sometimes this is the hardest decision of all!
If the x-ray is ABNORMAL, is the abnormality DIFFUSE or is it confined to discrete areas of the lung and therefore FOCAL?
- Furthermore, if the lung disease is DIFFUSE,
  - Is it basically AIR-SPACE disease (involving alveoli) or
  - Is it affecting mainly the supporting structures of the lung and, therefore, INTERSTITIAL?
In a computer-based algorithm, selections are made with single click of the mouse!

Suppose that the x-ray demonstrates DIFFUSE AIR-SPACE disease.

The algorithm then asks you if the patient's symptoms are ACUTE or CHRONIC.
- This question gives us crucial insight into the process of making radiographic diagnoses: One cannot, in general, make a reasonable list of possible diagnoses without knowing about the patient. Clinical history is vital.

Suppose that the patient's symptoms are ACUTE.

The algorithm now gives a list of possibilities rather than a single diagnosis.
- What a valuable insight! Although Radiology can "hit a home run" by coming up with the correct diagnosis when everyone else is baffled, this is not always--or even usually--the case.

Can we narrow the list of possibilities further?

The answer is "yes." We do this by ---
- Providing a more detailed clinical history.
  - For example, if the x-ray presents a "pulmonary edema" pattern but heart disease is not suspected, the algorithm suggests a different set of possible diagnoses.
- Performing a good physical examination.
- Requesting appropriate laboratory tests and other studies, such as EKG.
- Performing biopsies, when necessary.

Compared with neural networks, algorithmic approaches have advantages.

The algorithmic approach is transparent and easy to understand. No "black box" neural networks here!
Algorithms give a list of possible diagnoses rather than a single "best choice." This is usually more realistic in the "real world" in which not all possibly relevant data about the patient and the illness is known.

Compared with neural networks, algorithmic approaches also have disadvantages.

In an algorithmic approach, the designer has to think of every possibility.
- If your case has unusual features not covered by the algorithm, you have no place to go.
Algorithms tend to be inflexible, responding poorly to new situations. Neural networks provide solutions even when the specific parameters provided to them have never been encountered before.
When printed out on paper, algorithms can be confusing with arrows leading here, there, and everywhere. Also, there is only so much algorithmic information that can be included onto a single sheet of paper and still be readable. However, as we see in this presentation, computers can help overcome the problem of readability and provide clinical images as well.

revised -- January, 2000
revised -- December, 2002

Home -->

List of Projects -->