An expert system design is desired to obtain automated setup of all scan parameters for users with varied backgrounds, individualized to the examination requirements and consistent with overall scheduling considerations. A number of formidable obstacles must be overcome, relating both to performance and user interface. Potential performance obstacles relate to: (1) a combinatorial explosion which would result from consideration of all possible scan parameter settings, (21 varied examination requirements for different patients, (3) exceptional intrinsic flexibility of MR imagers, and (4) ideally, the desireability of a scan optimization approach. Potential user interface obstacles include: (1) widely varying user background and experience with MR techniques and (2) a strong subjective component to the process of scan setup.
A unique conceptual framework was developed to implement this expert system. The salient features of this expert system1e2 include (1) separation of the clinical and physical models, (2) identification of MR performance indices, (3) explicit consideration of tradeoffs for each performance index (4) option for user to bypass directly any part of the expert system,i.e., to set any scan parameters (~1 or tradeoffs explicitly. Eight performance indices are identified: (1) examination time, (2) lesion-to-background contrast, (3) signalto-noise ratio (S:N), (4) spatial resolution, (5) slice thickness, 16) survey length, (7) field of view, and (6) motion artifact rejection. Tradeoffs are expressed as priorities CX, and constraints (~1 for each of the performance indices. The system input from the user is a description of the clinical setting (x), obtained via interview, and/or direct settings (any combination of 2s or c's and/or E'S) specified by the user (see fig. 1). The system output is a set of optimized scan parameters given the clinical setting and constraints, or an indication that no solution is possible (if the problem is overconstrained).
The implementation, shown in fig. 1, consists of (1) a network of CONSULT-IR3expert systems, (2) a mathematical model and optimization algorithm, and (3) a user override option. As refinements, separate expert systems may be used (1) to generate estimated scan parameters which may improve convergence characteristics of the optimization algorithm and (2) for scheduling considerations.
The mathematical model consists of simplified equations that describe each of the performance indices in terms of scan parameters, eg., TR. number of views, number of excitations, hybrid echo planar factor, coil characteristics, motion rejection techniques, etc. The optimization procedure performs a process coranonly described as mathematical programming.
The advantages of this integrated expert system over conventional approaches relate to both performance and ease of use. The performance advantages are: (1) use of an explicit, testable model, (2) adaptability to an empirical expert system (CONSULT LEARNING SYSTEM TM3), (3) possibility of modular debugging of the subjective (expert system) and objective (mathematical model) components, (4) synthesis of stored and calculated operations which prevent combinatorial explosions and nonconvergence problems, and (51 adaptability to scheduling. Ease of use results from (1) a novel conceptual framework, (2) "anthropomorphic design" (which separates a "clinician' module-from a "physicist" module), (3) adaptability to a wide range of users, and (4) a user-friendly interface.