Hidden Markov Models and Surgery

I'm currently working on a project with the Electrical Engineering Department's Bioengineering Lab that will develop a way to quantify a surgeon's skill using Hidden Markov Models. My great thanks to Tim Kowalewski for telling me about this project and inviting me to contribute.

Currently, most surgeons' skills are qualitatively measured. An aspiring surgeon performs a sequence of operation tasks while a qualified surgeon looks on. The qualified surgeon rigorously, yet subjectively, determines whether or not the student can advance in ranks. In an attempt to make this process more quantitative, a team of engineers at the UW is designing a machine equipped with zillions of sensors that will attempt to measure the skill of a surgeon. At the heart of the current algorithm for doing so are hidden Markov models.

The idea is that several amateur, intermediate, and expert surgeons would complete various surgical tasks on this machine. The machine is equipped with various sensors that gather all sorts of data on the surgeon's movements, precision, stability, and other important factors. The data is used to create and train a hidden Markov model. Then, Dr. X comes into the lab and performs the same tasks on the machine. Their data is fed into the HMM where, given the initial "training" with known surgeon skill levels, the HMM can determine Dr. X's skill in relation to those previously known values. Hidden Markov models are being used in all sorts of settings such as image recognition, voice recognition and emulation, and even in the financial market --- actually, I helped William Stein incorporate an HMM module into the Sage Finance package.

My task for this project will be to link an open-source GHMM C library to the program the team is using to run the aforementioned machine. I'm also going to write some code that utilizes the library's routines; that is, parsing and sending the obtained surgical data to the appropriate GHMM modules and interpreting the output values. My role in the project is fairly open-ended so I might end up contributing more as time goes on.

Perhaps in my next post I'll give a little introduction on hidden Markov models!