COMPUTER-ASSISTED ASSESSMENT AND FEEDBACK FOR IMAGE-GUIDED INTERVENTIONS TRAINING
To improve patient safety, many surgical and interventional specialties use simulation training. This ensures that trainees have a minimum level of proficiency prior to real patient encounters. During simulation training, it is imperative to assess trainees’ learning curves and instruct trainees how to improve their skills. It is infeasible, however, for expert preceptors to continually monitor trainees during training. As a supplement to expert supervision, computer-assisted training has emerged. This thesis proposes computer-assisted methods for assessment, feedback, and instruction in image-guided interventions training. This thesis makes the following contributions. 1. A comprehensive and methodical review of the literature in computer-assisted assessment methods for image-guided interventions training. The review outlines characteristics by which assessment methods can be classified and identifies tools available to the community. 2. Studies on the development and validation of performance metrics for skills assessment in image-guided interventions. The analysis of hand and arm movements in simulated colonoscopy distinguished between novice and experts. The quantification of the planning process in deep brain stimulation indicated skill and was useful for monitoring learning curves during a training course. In both cases, application-specific metrics formulated in consultation with clinical experts had added value in assessment. 3. Methods for overall skills assessment in ultrasound-guided interventions based on performance metrics. A framework for determining which aspects of an intervention are assessed by which performance metrics was designed and validated in three applications. A method for overall skills assessment using transparent and configurable machine learning was developed. It was shown to both be accurate and provide useful feedback to trainees automatically. 4. Development of a method for automated workflow analysis in ultrasound-guided needle interventions. The method combines domain-knowledge with pattern recognition techniques to provide instruction in real-time. The method achieves accuracy approaching the accuracy attained by human observers. Automated assessment, feedback, and instruction is feasible for computer-assisted training in image-guided interventions when an expert preceptor cannot be present. In this context, the proposed methods have been developed and validated in real training scenarios. The methods are made available to the community through Perk Tutor (www.perktutor.org), an open-source image-guided interventions training platform.