Research On Application Of Sequential Mining In Discovery Of Clinical Behavior Patterns

Nowadays, the medical institutions are facing these hotspot issues as to improve the standardization of quality management in medical treatment and to prevent medical resource abuses and nonstandard treatment process. In medical field, the clinical pathway management theory, was proposed to instruct medical treatment, which made strict working sequence, accurate time requirement/limitation for the diagnosis and treat process with expectation to normalize treatment behaviors and medical resource using, and promote medical quality, aiming at the monitoring, treatment, rehabilitation and nursing care of the single disease.According to the instructing principles of Evidence-based medicine, making a good clinical treatment solution needs to resort to the objectively reliable data and criteria and some concrete evaluation methods, and in the mean time comply with the scientific digital measure standard in analyzing the medical irregularities. As such, this dissertation applies data mining technologies to the existent medical data to discover the hidden but legal clinical behavior pattern structures and fragments by building data processing model and performing sequential mining on the data, in order to provide decision making basis for the medical treatment behavior management and irregularities detection. These researches will bring crucial theoretical and practical values. The main works of the dissertation are listed as follows:1. Studies on data preprocessing methods. Operations as attribute reduction, extraction and cleaning are performed on the raw data to obtain valid clinical behavior data. Because the cleaned data still has the time relationship noises, and when directly applying sequential mining algorithm on them, it is hard to discover the high quality patterns. So, a time normalization model that defines the ordinal and parallel relationships in the temporal behaviors is used to compute the relationship intersection coefficient. Based on the calculated results to identify the noises in time relationship of the temporal behaviors, and comply with the rule, i.e. no noises and keeping the originally correct relationships between the behaviors to carry out noise cleanup.2. Since the data mining algorithms have characteristics as special requirements of format and domain behavior, instead of the fixed window algorithm, a sequential generation algorithm with small granularity and duration constraint is proposed, which allows the clinical behavior to be triggered across windows, brings compound of ordinal and parallel patterns and generates the sequences that can be directly used to be mined.3. A sequence mining method with duration constraint abbreviated SMDC is proposed based on the combination of Apriori and Spam algorithms and clinical behavior characteristics with respect to the weak points of present algorithms. By adopting incomplete trie storage structure, the candidate patterns can be generated by extending the original patterns in the standard DFS manner, meanwhile, seeds of new nodes are built dynamically. And the following is how the support is computed through using the candidate set selection with duration constraint method: introduce double supports and binary operation flag to compute the subsequence; introduce node legality concept in order for using duration constraint to identify the legality of subsequence, in terms of describing relationships between the legal and illegal nodes and their functionalities, the correct candidate items are assured to be continually generated and the legal subsequence support which conforms to clinical behavior standard is calculated out, achieving pruning the tree nodes.4. Verifying the proposed data mining algorithm and implementing the prototype of clinical behavior pattern mining to dig the clinical data processed by relevant preprocessing algorithm, and by analyzing and evaluating the patterns to present their legitimacy and usability.