Application of image processing and statistical analysis to model patient outcomes in cochlear implantation and epilepsy surgery
The analysis of image data in image guided surgery can produce promising insights to improve patient outcomes, since medical images provide detailed, patient-specific data on how an intervention affects the anatomy – or in the case of an interventional device, how it relates to the anatomy. However, there are also significant technical challenges in such analyses, largely because the selection and validation of image registration, segmentation, and modeling techniques depend on the study and the data available. In this dissertation, we discuss building image processing pipelines to extract position information and applying statistical methods to analyze patient outcomes in two surgical interventions – cochlear implantation and laser interstitial thermal therapy (LITT) for mesial temporal lobe epilepsy (mTLE). Cochlear implants are neuroprosthetics for moderate to profound hearing loss, and implanting the electrode array in the cochlea is a critical surgical component. LITT is a thermoablative surgery where a laser catheter is inserted up to a chosen target area around the hippocampus to ablate the seizure focus for mTLE patients. In both cases, there is evidence that the position of the array or the target location for ablation affects outcomes, without strong consensus on what constitutes good positioning. We build separate datasets for the two applications using preoperative and postoperative images as well as relevant clinical data. We select and validate appropriate image processing methods and extract positioning information from image datasets using semi-supervised algorithms. Using linear regression modeling and conditional probability mapping, we analyze the relationship of the outcome data with the surgical image data and identify the positioning necessary for good outcomes. The major contributions of the dissertation include insights for future surgeries to obtain better patient outcomes, as well as the novel image processing and statistical analysis pipelines we created to analyze the correlation between patient outcomes and the anatomical position of intervention.