inverse problem

After going over a machine learning paper with a friend, I realized that clinic is just solving the inverse problem from biology.  With biology and pathophysiology, we start with some kind of problem and trace the course out to the disease; with clinic, we assemble a list of the salient features from the history and exam and identify relevant patterns.  We also try to think about the prior probability that the patient has the disease and also what the prevalence is in the population - all Bayes' Theorem.

With that in mind, I'll be going into extract the relevant information (hopefully in a  nice, appropriate conversation with history-taking and a physical exam), format it into the desired format (the write-up or oral patient presentation), and then the attending physician will perform inference and pattern recognition (having already figured out what features are most predictive from training) to determine what tests need to be performed, if any, and then we will discuss the management and treatment of disease.

...Except that I actually have to learn each line in the table of diseases and features myself - no computer to help.  Learning seems to haphazard and the inference rather squishy.  We'll see how this goes.

Knowing that clinical medicine is just learning to solve the inverse problem makes me feel much better.  I never realized that inference was what we were training to do.

Here goes!