Rapid Communication Wavelet analysis for detecting body-movement artifacts in optical topography signals Hiroki Sato, a,b, ⁎ Naoki Tanaka, a Mariko Uchida, b Yukiko Hirabayashi, a,b Makoto Kanai, c Takashi Ashida, c Ikuo Konishi, c and Atsushi Maki a,b a Advanced Research Laboratory, 2520 Akanuma, Hatoyama, Saitama 350-0395, Japan b JST (Japan Science and Technology Agency)/CREST (Core Research for Evolutional Science and Technology), Japan c Department of Obstetrics and Gynecology, Shinshu University School of Medicine, Japan Received 30 October 2005; revised 9 June 2006; accepted 18 June 2006 Available online 28 August 2006 We have developed a wavelet-based method of detecting body- movement artifacts in optical topography (OT) signals. Although OT, which is a noninvasive imaging technique for measuring hemodynamic response related to brain activation, is particularly useful for studying infants, the signals occasionally contain undesirable artifacts caused by body movements, so data corrupted by body- movement artifacts must be eliminated to obtain reliable results. For this purpose, we applied a wavelet transform to automatically detect body-movement artifacts in OT signals. We measured OT signals from nine healthy infants in response to speech stimuli. After the continuous signals had been divided into blocks (a block is a time series of OT signal in a 30-s period including a 10-s stimulation period), they were classified into two groups (movement blocks and non-movement blocks) according to whether the participants moved or not by video judgment. Using those data, we developed a wavelet-based algorithm for detecting body-movement artifacts at a high discrimination rate being consistent with the actual body-movement state. The wavelet method has two parameters (scale and threshold), and a Monte Carlo analysis gave the mean optimal parameters as 9 ± 1.9 (mean ± standard deviation) for the scale and as 42.7 ± 1.9 for the threshold. Our wavelet method with the mean optimal parameters (scale = 9, threshold = 43) achieved a higher discrimination rate (mean ± standard deviation: 86.3 ± 8.8%) for actual body movement than a previous method (mean ± standard deviation: 80.6 ± 8.7%) among different partici- pants (paired t test: t(8) = 2.92, p < 0.05). These results demonstrate that our wavelet method is useful in practice for eliminating blocks containing body-movement artifacts in OT signals. It will contribute to obtaining reliable results from OT studies of infants. © 2006 Elsevier Inc. All rights reserved. Keywords: Optical topography; Body-movement artifact; Wavelet; Infant; Brain function; Near-infrared spectroscopy (NIRS) Introduction Near-infrared spectroscopy (NIRS) has been used for non- invasive measurements of concentration changes in oxygenated hemoglobin (oxy-Hb), deoxygenated hemoglobin (deoxy-Hb), and total hemoglobin (total-Hb) related to brain functions (Chance et al., 1993; Hoshi and Tamura, 1993; Kato et al., 1993; Villringer et al., 1993). Optical topography (OT) is an application of NIRS using multiple measurement positions, which allows brain activation to be imaged (Koizumi et al., 1999; Maki et al., 1995). OT is uniquely useful because it is noninvasive and can be used without restraining the participant, so it can measure brain functions in healthy infants (Pena et al., 2003; Taga et al., 2003). However, OT has the drawback that fast head movements by participants can cause undesirable noise (body-movement artifacts) in the OT signals (oxy-Hb, deoxy-Hb, and total-Hb signals). This is a serious problem in studying healthy infants because they move even in their natural sleep, and they cannot be instructed to keep still. One solution is to make an artifact-free probe holder (cap) that perfectly fits a participant’ s head and reduces gross movement artifacts (Aslin and Mehler, 2005). However, no faultless probe caps have been developed yet as far as we know. Therefore, most researchers have taken an analytical solution to the problem: body- movement artifacts are detected and the corrupt blocks are removed from the analysis (Homae et al., 2006; Kotilahti et al., 2005; Pena et al., 2003; Taga et al., 2003; Taga et al., 2000; Wilcox et al., 2005). However, the descriptions of the criteria for finding movement artifacts are not good enough to follow, for example, ‘movement artifacts resulted in sharp spikes’ (Taga et al., 2000), ‘movement artifacts, which were detected by the analysis of sharp changes in the time series’ (Taga et al., 2003), and ‘movement artifacts’ (Homae et al., 2006). Other studies described more quantitative approaches; for example, changes larger than 0.1 mM·mm over two successive samples (during 200 ms) were regarded as artifacts in filtered (with a bandpass filter between 0.02 and 1 Hz) signals (Pena et al., 2003), a peak-to-peak threshold of www.elsevier.com/locate/ynimg NeuroImage 33 (2006) 580 – 587 ⁎ Corresponding author. Advanced Research Laboratory, Hitachi, Ltd., 2520 Akanuma, Hatoyama, Saitama 350-0395, Japan. Fax: +81 49 296 5999. E-mail address: hiroki.sato.ry@hitachi.com (H. Sato). Available online on ScienceDirect (www.sciencedirect.com). 1053-8119/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2006.06.028