Eliciting Tacit Expertise in 3D Volume Segmentation

Abstract

The output of 3D volume segmentation is crucial to a wide range of endeavors. Producing accurate segmentations often proves to be both inefficient and challenging, in part due to lack of imaging data quality (contrast and resolution), and because of ambiguity in the data that can only be resolved with higher-level knowledge of the structure and the context wherein it resides. Automatic and semi-automatic approaches are improving, but in many cases still fail or require substantial manual clean-up or intervention. Expert manual segmentation and review is therefore still the gold standard for many applications. Unfortunately, existing tools (both custom-made and commercial) are often designed based on the underlying algorithm, not the best method for expressing higher-level intention. Our goal is to analyze manual (or semi-automatic) segmentation to gain a better understanding of both low-level (perceptual tasks and actions) and high-level decision making. This can be used to produce segmentation tools that are more accurate, efficient, and easier to use. Questioning or observation alone is insufficient to capture this information, so we utilize a hybrid capture protocol that blends observation, surveys, and eye tracking. We then developed, and validated, data coding schemes capable of discerning low-level actions and overall task structures.

A Summer Research Experience in Robotics

“…As indirect evaluation, several of the projects from the first two summers have been published (Hubicki et al 2016;Bowen-Biggs et al 2016;John et al 2016;Hubers et al 2015a;2015b;Holloway et al 2015;West et al 2016;Cesare et al 2015;Curran et al 2015;Faraji et al 2015;Allani et al 2016;Faraji et al 2015;Sundberg et al 2016), with the ones from this summer in preparation.…”

Developing and Validating a Computer-Based Training Tool for Inferring 2D Cross-Sections of Complex 3D Structures

“…Results from field studies [47,62] suggest that expert segmenters have a better understanding of the expected 2D cross-sections of 3D structures. To conduct an accurate segmentation, it is not sufficient to just passively view what a 3D structure and its cross-section look like; expert segmenters also actively search for cross-sections that meet certain criteria [28].…”

“…To choose the tasks, we began by defining a set of canonical shape types (based on ones seen in the observational studies [47,62]) and unique, identifiable cross-sections that arise from those shapes. We then chose four 3D objects to cover a range of object complexity: Two simple objects (an hourglass and a tapering shape), and two complex objects (branching Y shape and a potato shape with a hole).…”

“…Also, previous studies show it is difficult to infer cross-sections of complex 3D structures without user control over the viewpoints and objects [28,47,62]. To our knowledge, there is no training tool--let alone a computer-based one--to support user control and interaction in this context.…”

“…For example, in medical imaging (in particular, 3D volume segmentation), the segmenter operates on 2D cross-sections of the 3D data, mentally integrating contours drawn on these slices into a coherent 3D anatomical structure. Results from in-depth field studies on segmentation with human experts indicate that they have a better understanding of the expected 2D cross-sections of 3D structures—and tend to use 3D views more frequently—than novices (Sanandaji et al, 2016; West et al, 2016). Still, some of the more repetitive and time-intensive tasks of manual segmentation are usually assigned to novices (e.g., students assistants).…”