- 研究番号：25C10
- 研究分野：interdisciplinary
- 研究種別：奨励研究
- 研究期間：2025年04月〜2026年03月
代表研究者
パタショヴ デミテリ 理工総研が募集する次席研究員
PATASHOV Dmitry Junior Researcher
理工学術院総合研究所 酒井 弘 研究室
Waseda Research Institute for Science and Engineering
URL：https://w-rdb.waseda.jp/html/100003379_ja.html
研究概要
1) Development of MEG data processing and analysis methods for neural decoding
2) Development of fNIRS data processing and analysis methods
2a) For language development study in infants.
2b) For the assessment of brain disorders.
The purpose of these studies is to develop robust and reliable methods that allow for a more accurate analysis of MEG and fNIRS data than those produced by the currently available techniques. This in turn will help us to better understand the mechanisms of speech production, language development, effects of brain disorders and more. Understanding the underlying mechanisms is essential for many real-life applications. For example, advances in neural decoding may help to communicate with people suffering from “locked-in syndrome”. Understanding the language development process, may help to detect developmental disorders at an early stage. Accurate analysis of brain disorder’s effects may help in the design of new, more accurate and/or reliable diagnosis techniques, as well as in the design of treatment procedures. Since the accuracy of the results of subject assessments heavily rely on the methodology used to analyze the data, it is important to use sophisticated methods that provide maximum precision
When working with Neuroimaging data, it is important to understand not only its biological factors, but also the statistical nature of the recorded stochastic processes. Biological factors, although important and are mandatory to take into consideration when analyzing these types of data, by themselves are usually insufficient to achieve stable, reliable and high accuracy results. Unfortunately, other important factors, such as physics involved in the mechanisms of the recording devices, the stochastic nature of the recorded data or the limitations and underlying assumptions of the engineering techniques used in the analysis, are often overlooked. Our goal is to collaborate with experts from different fields in solving of these highly multidisciplinary research questions. We aim to utilize methods that suit all the limitations and constrains of the recorded data. For example, one of the commonly overlooked issues of the neuroimaging data is its non-stationary nature. Most of the classical engineering solutions for the filtration of the data, have an underlying assumption that the data is at least wide-sense stationary. Nonetheless, most publication in this field, do not address this issue in any way. In our approach, we are examining the quality of processing and analysis techniques that do not assume linear behavior or stationarity of the data. We strive to find the best matching solutions that may involve non-stationary methods only or a well thought through combinations of stationary techniques together with non-stationary ones in order to minimize data distortions and warping of the assessments results. To serve as an example, we utilize methods such as Hilber Huang Transform, Empirical Mode Decomposition, Variational Mode Decomposition, Cumulative Curve Fitting Approximation, Riemannian Manifold, Graph Signal Processing and other sophisticated techniques, both by themselves and in combination with classical approaches. We compare these advanced methodologies to the ones widely used today to find solutions with best performance.
