Analysis of similarity measure in the longitudinal study using improved longest common subsequence method for lung cancer. (January 2015)

• Record Type: Journal Article
• Authors: Fang, LiYing; Wan, Min; Yu, MingWei; Yan, JianZhuo; Liu, Zheng; Wang, Pu
• Is Part Of: Biomedical signal processing and control. Volume 15(2015)
• Page Start: 60
• Page End: 66
• Publication Date: 2015-01
• Subjects: Longitudinal data -- Time-series -- Similarity measure -- Longest common subsequence -- Hierarchical clustering; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2014.09.010
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: We examine the similarity between clinical time-series categorical data. Increasing constraint window parameter will decrease the model performance. The proposed W-LCSS-CW method performed significantly better than other three methods based on the relative evaluation. The proposed W-LCSS-CW method performed significantly better than other three methods based on the external evaluation. The constraint window and the weight factor is useful in measuring sequence similarity. Abstract: Background: In clinical practice, longitudinal data can be used to find trend patterns of pathema progress, such as tumour progress, along a time axis. This kind of data can be treated as time-series data. The maximum common sub-sequence is the most common method for calculating similarity of time-series data; and each point is normally treated as having the same weight. However, not all points of data within the time series should be given the same importance. According to clinical experience, the later period sub-sequence (closer to death) has a more significant effect than earlier periods in a trend analysis. Results: A weighted-similarity measure based on LCSS with Constraint Window (W-LCSS-CW) Method is proposed. The results obtained from the time-series data using different weighting factors are discussed. In a study of non-small cell lung cancer using time-series data, the relative evaluation method and external evaluation method were adopted to calculate cluster effect. The results show that the proposed method, W-LCSS-CW, can improve clustering performance significantly. Clustering performance of various methods was performed using a comparison of ( C index / M index ). The proposed W-LCSS-CW Method was evaluated to 1.55 which was 37.02%, 48.01%, 49.64% higher than other common methods (Euclidean, DTW, STS) respectively. Conclusions: The proposed W-LCSS-CW Method is recommended for monitoring time-series data of tumour patients because the incorporated weighting factor provides more convincing cluster results for medical assist support.