Image understanding system for carotid angiograms

摘要

The paper describes a prototype image understanding system for carotid angiograms. The system currently accepts and analyses binary images of vessels in the vicinity of the carotid bifurcation. A knowledge base consisting of approximately 200 rules is used to reach a consistent interpretation and labelling of these vessels. This knowledge base contains both facts (domain) knowledge which is drawn from various areas of science and strategy (control) knowledge which determines how and when the facts knowledge should be used. The knowledge base contains knowledge (both facts and strategy) relating to image preprocessing, image structuring, general vascular anatomy and regional anatomy. These various tvpes of knowledge are partitioned so that the system could, in principle, be easily adapted for use in other application areas. The system is expectation driven (model driven) with the inference engine using a backward chaining mechanism and a blackboard during the analysis of an angiogram. A crude user interface gives a running commentary on the current goals and eventually provides details of the final interpretation and labelling of the vessels in the angiogram. The system is implemented using POP-11 and PASCAL and runs on a Vax 11/750. Twelve binary images of vessels in real angiograms have been analysed and labelled correctly bv the system.