Towards the automatic control of laser ablation for surgical applications
The goal of this thesis is to propose and investigate a method of predicting depth of a laser dissection pulse in soft tissue without acquiring material properties of the tissue target or measuring the laser output. The method proposed is similar to what is used by laser surgical operators today, but uses regression learning to perform on-the fly predictions in place of a skilled laser surgeon. Power of the laser and the ablation depth were recorded for 57 samples and fed into the regression algorithm. Data exclusion was performed using Temperature before laser action as criteria. A linear and logarithmic model was explored using random points from the data post-exclusion, validation RMSE ranged from 135-200 micrometer. A linear and logarithmic model was explored using data points below a moving power threshold and validated with data points above said threshold, validation RMSE ranged from 108-170 micrometer. The t.test performed showed there was not a significant difference between the linear and the logarithmic models' goodness of fit metrics, but it did show there was a significant difference between the model building methods (randomly selected data points, moving power threshold). The method of building a model using lower power levels to predict larger power levels had better goodness of fit metrics than the method of selecting data points at random. In the future, this method could be used to help approximate the laser settings for surgery on a procedural basis, and allow for surgeons to perform at a higher skill level with less training.