ترکیب دانش تخصصی و داده کاوی در یک دامنه تشخیص پزشکی

The medical diagnosis system described here uses underlying knowledge in the isokinetic domain, obtained by combining the expertise of a physician specialised in isokinetic techniques and data mining techniques applied to a set of existing data. An isokinetic machine is basically a physical support on which patients exercise one of their joints, in this case the knee, according to different ranges of movement and at a constant speed. The data on muscle strength supplied by the machine are processed by an expert system that has built-in knowledge elicited from an expert in isokinetics. It cleans and pre-processes the data and conducts an intelligent analysis of the parameters and morphology of the isokinetic curves. Data mining methods based on the discovery of sequential patterns in time series and the fast Fourier transform, which identifies similarities and differences among exercises, were applied to the processed information to characterise injuries and discover reference patterns specific to populations. The results obtained were applied in two environments: one for the blind and another for elite athletes.

This paper shows the results of the I4 project1 (Intelligent Interpretation of Isokinetic Information). It describes a medical diagnosis system in the field of physiotherapy and, more specifically, muscle function assessment based on isokinetic machine data, using an expert system and data mining techniques. An isokinetic machine can be described as apparatus on which patients perform strength exercises. This machine has the peculiarity of limiting the range of movement and the intensity of effort at a constant speed (which explains the term isokinetic). Data concerning the strength exerted by the patient throughout the exercise are recorded and stored in the machine so that physicians can visually analyse the results using specialised computer software. The information supplied by an isokinetics machine has a lot of potential uses (López-Illescas, 1993): muscular diagnosis and rehabilitation, injury prevention, training evaluation and planning, etc. However, the software built into these systems, and even the isokinetic-based diagnosis techniques themselves, still have some significant handicaps that have detracted from the success of this field: • Standard software provides only an analogical representation of the massive data flow output by these systems. The physician is left to analyse this with no further help. This is not an easy task, as it depends almost exclusively on the personal experience of the therapist. • Novice therapists find it enormously difficult to interpret and understand the output graphs. • Decisions are guided by the therapist's instinct, as there are no models that can be used as a reference for most of the common injuries. Moreover, the few simple models that do exist have merely been stated by experts and are not founded on rigorous data analysis. However, there is a huge amount of stored information (performed tests) that has not been analysed to improve the procedure as a whole. Due to the above-mentioned problems, system design should combine both practitioner expertise and knowledge that can be discovered within the data. Three objectives were therefore defined: (a) Equip the isokinetic system with a knowledge-based system (KBS) that would perform an intelligent analysis of the strength curves output by the isokinetic test on which the assessments are based, modelling the knowledge of the expert who works the machine. This would be a valuable aid for the examining physician in detecting possible injuries. (b) Characterise injuries. Bearing in mind that a huge number of isokinetic tests had already been performed and stored in a database, we aimed to find any sort of patterns useful for characterising different injuries in terms of isokinetic data. These patterns would be extremely valuable in two ways: as a useful research tool for therapists adding to the knowledge about the isokinetic shapes of common injuries and as reference models to be used for injury classification and, if possible, diagnosis. Data mining techniques based on the discovery of sequential patterns in time series were applied for this purpose. (c) Create reference models. The third objective involved discovering standard patterns that characterised specific population types taken from the isokinetic data already prepared and stored in the database. For example, the process of evaluating a particular athlete against a standard curve, specific to his or her sport, is a very effective means for assessing athletes' capacity and potential for the sport they intend to go in for.

The development of an ES and its later refinement is mainly based on eliciting and entering experts' knowledge of the subject into the system. Thanks to data mining techniques, a more efficient and objective process for developing an ES can be applied to complement the above, provided enough data are available from which new knowledge can be discovered. The I4 project described here is an example of this approach applied to the expert processing of isokinetic data. Initially, the expert knowledge of the isokinetic physician was entered into the system so as to conduct an intelligent analysis of the numerical parameters and morphology of the strength curves output by the isokinetic tests. Considering the volume of tests, data mining techniques, involving processing time series to discover patterns, injuries and reference models, were then applied. The above expert system plays an active role in some of the data preparation and cleaning tasks of this knowledge discovery process (eliminating incorrect tests, exercises or parts of them and eliminating inertia peaks from the strength curves). After evaluation and validation by the expert, this new knowledge was entered into the expert system, which performed better and was more efficient than the one directly elicited from the expert. The expert participation in the KDD process is not new. Indeed, it is important in most of these processes to have an expert who is familiar with the data to deal with data cleaning, pre-processing and evaluation. There are also several examples of co-operation in these two areas, for example, building a KBS from the discovered rules (Lee et al., 2001 and Tsumoto, 1999). However, the originality of the proposal described in this paper lies in the fact that an ES was built that directly intervenes in the KDD process. Once this process is complete, the discovered knowledge can be fed back to the ES. It is noteworthy that the participation of experts throughout the entire KDD process was fundamental, as a vital source of knowledge and also to get them to identify with the project, thus overcoming their traditional reticence concerning technologies of this sort.