Spatiotemporal Analysis of Swallowing Movement via 4DCT4DCTを用いた嚥下運動の時空間分析

Focus研究焦点

Spatiotemporal Analysis of Swallowing Movement via 4DCT4DCTを用いた嚥下運動の時空間分析

This project focuses on the spatiotemporal analysis of swallowing mechanics using four-dimensional computed tomography (4DCT), a dynamic imaging modality that acquires volumetric CT data at multiple successive time points to capture the motion of anatomical structures over time. Swallowing is a highly coordinated neuromuscular process involving the simultaneous movement of dozens of structures including the tongue, hyoid bone, larynx, pharynx, and esophagus. Dysphagia — impaired swallowing — is a clinically important condition associated with aspiration pneumonia, malnutrition, and reduced quality of life, particularly in elderly and neurologically impaired populations. A precise, quantitative understanding of swallowing kinematics is therefore essential both for clinical diagnosis and for the development of targeted rehabilitation strategies.

The core technical challenge is that 4DCT produces sequences of three-dimensional volumes at fine temporal resolution — in this case, frames captured every 0.1 seconds up to 0.4 seconds — resulting in a large, complex dataset that requires automated analysis pipelines to extract clinically meaningful measurements. The proposed pipeline consists of two main stages. In the first stage, a deep learning-based segmentation network is applied independently to each temporal frame of the 4DCT sequence. This network segments the relevant swallowing structures — including the pharynx, hyoid, and surrounding soft tissues — and produces labeled volumetric masks for each time point. The use of an automated segmentation approach is critical, as manual delineation of these structures across multiple frames is prohibitively time-consuming.

Once segmentations are obtained across all time points, the second stage of the analysis focuses on quantifying the dynamics of the pharyngeal cavity over time. A key metric analyzed in this project is the cross-sectional area profile of the pharyngeal lumen along its longitudinal axis. This is computed by resampling the segmented pharyngeal region along a centerline defined from the segmentation, extracting cross-sectional slices at regular intervals, and measuring the area of each slice as a function of both position along the pharynx and time. The resulting data forms a spatiotemporal map — a two-dimensional representation with position on one axis and time on the other — which directly visualizes the wave-like contraction of the pharyngeal walls during swallowing.

Two primary spatiotemporal maps are computed and analyzed: one characterizing pharyngeal contraction dynamics, and another characterizing bolus passage through the pharynx. The pharyngeal contraction map reflects the mechanical squeezing action of the pharyngeal walls, which propels the food bolus toward the esophagus. The bolus passage map, in contrast, captures the spatial and temporal distribution of the food material as it moves through the pharyngeal cavity. Together, these two maps provide a comprehensive and complementary description of swallowing function, allowing quantitative comparison between different swallowing conditions, between patients with and without dysphagia, or before and after therapeutic interventions.

This approach offers several significant advantages over existing methods for swallowing evaluation, such as videofluoroscopic swallowing studies (VFSS) or fiberoptic endoscopic evaluation. Unlike these methods, which rely on 2D projections or direct visual inspection, 4DCT-based analysis provides fully three-dimensional, quantitative data throughout the entire swallowing event. The automated segmentation and spatiotemporal mapping pipeline reduces the burden of manual analysis and enables reproducible measurements. Challenges include the relatively high radiation dose associated with 4DCT acquisition and the computational cost of processing large volumetric sequences, both of which are active areas of development in the field.

本研究は、四次元CT（4DCT）を用いた嚥下動態の時空間分析に取り組んでいます。4DCTは連続した複数の時点で体積CTデータを取得する動態撮像モダリティであり、解剖学的構造の動きを経時的に捉えることができます。嚥下は、舌・舌骨・喉頭・咽頭・食道など多数の構造が関与する高度に協調されたの神経筋プロセスです。嚥下障害（嚥下困難）は誤嚥性肺炎・栄養不良・QOL低下と関連する臨床的に重要な状態であり、特に高齢者や神経障害患者に多く見られます。

中核的な技術的課題は、4DCTが0.1秒ごとに三次元体積を収集することで生成される大規模・複雑なデータセットの自動解析です。提案パイプラインは二段階で構成されます。第一段階では、深層学習に基づくセグメンテーションネットワークを4DCTシーケンスの各フレームに独立して適用し、咽頭・舌骨・周辺軟部組織などの嚥下関連構造をラベル付きボリュームマスクとして抽出します。複数フレームにわたる手動輪郭描出は非常に時間がかかるため、自動セグメンテーションの活用が不可欠です。

全時点のセグメンテーション取得後、第二段階では咽頭腔の動態の定量化に焦点を当てます。主要な評価指標は、咽頭腔の長軸方向の断面積プロファイルです。セグメンテーションから定義された中心軸に沿って咽頭領域をリサンプリングし、一定間隔で断面スライスを抽出して各位置での面積を計測します。得られたデータは、咽頭の位置を縦軸に、時間を横軸にした時空間マップを形成し、嚥下時の咽頭壁の波状収縮を直接可視化します。

咽頭収縮ダイナミクスを示すマップと食塊通過を示すマップの2種類が算出・解析されます。咽頭収縮マップは咽頭壁の機械的な絞り込み動作を反映し、食塊を食道方向へ推進します。食塊通過マップは食材が咽頭腔を移動する際の空間的・時間的分布を捉えます。この二つのマップにより、嚥下機能の包括的・補完的な記述が可能となり、異なる嚥下条件間・嚥下障害の有無・治療介入前後の定量的比較が実現します。

本手法は、2D投影や直接視覚観察に依存する既存の嚥下評価法（VFSSや内視鏡的評価）と比較して、嚥下全過程にわたる完全三次元・定量的データを提供するという点で大きな利点を持ちます。自動セグメンテーションと時空間マッピングパイプラインにより、手動分析の負担が軽減され再現性のある計測が可能となります。課題としては、4DCT撮影に伴う比較的高い被ばく線量と大規模体積シーケンス処理の計算コストが挙げられ、いずれも活発な研究開発が続いています。



Project 5
Development of an Automated Diagnostic System for Osteonecrosis of the Femoral Head Classification Based on MRI Images
MRI画像から複数種類の病型分類に基づく大腿骨頭壊死症診断を行う自動システムの構築
Automated Classification
MRI Segmentation
Osteonecrosis
Multi-scheme Diagnosis
Deep Learning