Automatic Segmentation: Some Ideas

During the last meeting, I was asked if I had any ideas on how to approach the problem of automatically segmenting the different features of the heart. While I currently lack the background to speak definitively about the "low level" technical details of creating such an algorithm, I have several usability ideas that pertain to how the program will eventually be used by the clinician. I believe that, while the accuracy of the algorithm is certainly an extremely aspect of the program's design, the interface itself must be intuitive enough to be used easily while conforming to the user's expectations in operation and implementation. In many ways, the interface itself is just as important as the algorithm behind it.

1. Users like control
Currently, the algorithm developed by Noel Codella is highly automated, but in doing so, takes much of the control away from the user. To operate the program, the user must simply identify the left ventricle in one slice of the cine MRI with a single click. The algorithm handles the rest. But to an experienced clinician who is used to performing all of the tasks manually, this degree of automation is a little disconcerting. I believe adding an option for the user to "tweak" the algorithm's segmentation decisions would lessen this sense of detatchment. While it is debatable whether a computer or a human can provide the best segmentation decisions, giving the option to perform manual corrections would make the user far more comfortable in using the program.

2. Users like feedback
As much as everyone loves to hate the little status bars the creep across the screen while we perform tasks on the computer, they are essential. I never really realized this until I saw the automatic segmenter in action. The program, while relatively fast, requires up to 30 to 45 seconds to fully complete segmentation after the initial user input. Unfortunately, there is no visual indication concerning the program's progress during this time. In every trial thus far, the program has finished successfully every single time it has been run, so it would seem that a status bar would serve little purpose other than to provide a little visual entertainment for the user. In reality, I've found that this has been one of the major complaints about the program's current implementation. Users like feedback about what the computer is doing!

These are all pretty minor suggestions that have little to nothing to do with the algorithm itself, but I feel changes such as these would make the program far more user friendly and will make it more likely for the program to be used in an actual clinical setting in the future.