The Craniofacial Research Instrumentation Laboratory (CRIL)
at the University of California, San Francisco, has been
investigating the application of image Processing techniques
to the Problem of feature location and tracking on
biostereometric images /2/. "An array camera was interfaced
to a small computer and a System developed for image
acquisition and analysis, and feature extraction. Previous
Work has included the development of Procedures for array
Camera calibration /3/. This Paper contains a report on a
series of experiments on feature extraction and tracking
conducted with Sequential digital X-ray images.
2. PROBLEM DEFINITION
There are two major categories of Points that need to be
located and tracked on biostereometric images: discrete
markers and "fuzzy" anatomic landmarks. The following
discussion will be concerned Primarily with the tracking of
anatomic features on a series of time-separated two
dimensional Cranial X-rays.
The CRIL physical records data base Presently contains a
number of longitudinal data sets, Covering hundreds of
patients who have undergone various forms of orthodontic
treatment. Each patient's physical records include
treatment records, study casts, and standardized Cranial x-
rays taken at regular intervals, Some patients may have as
many as ten films, covering up to eight Years, including
Pre-treatment ang post-treatment X-rays. In order to study
the physical effects of different orthodontic treatment
Programs, a number of significant anatomic landmarks are
identified and digitized on each film. For example, two
standard points (sella and nasion) are chosen to define a
frame of reference for the anatomic coordinate System. In
addition, points defining tooth locations and orientations,
the palatal plane, and mandibular and maxillary Structures
are digitized. The same Series of points is located on each
film of a patient, and the Process repeated for each patient
in a study. Eventually, à data base of Coordinate
information is Constructed, from which points can be
extracted, geometric relationships among points computed,
and biological hypotheses tested,
The process of anatomic landmark location is complex, X-ray
films Vary in contrast and intensity. Anatomic feature
Shapes and sizes change from patient to patient, and vary
within a single patient over time. The current manual
methods draw upon the experience of the investigator to
Overcome the Uncertainties resulting from these factors,
Thus, the automation of the location task faces significant
difficulties.
It was decided that any attempt to completely automate the
Process would not be appropriate due to the complexities
discussed above, Instead, the initial approach would focus
on the point tracking Problem, and would use a human
Operator to assist in point identification.
- 164 -