CN-121545201-B - Pain assessment system for neonatal facial expression image analysis

CN121545201BCN 121545201 BCN121545201 BCN 121545201BCN-121545201-B

Abstract

The invention relates to the technical field of medical image processing, in particular to a pain assessment system for neonatal facial expression image analysis, which comprises a data acquisition module, a facial expression manifold construction module, a curvature field analysis module, a space-time feature level analysis module and a pain assessment and alarm module, wherein the facial expression manifold construction module maps discrete facial feature points to a Riemann manifold space to construct an individualized expression manifold model; the invention remarkably improves the capturing sensitivity of the micro-expression, has robustness to shooting condition change, improves the distinguishing capability of different types of pains, realizes objective and quantitative evaluation and multistage early warning of the neonatal pain state, and provides reliable technical support for clinical pain management.

Inventors

ZHANG WENYUE
WANG XUEHUI
WANG LE

Assignees

中国人民解放军总医院第七医学中心

Dates

Publication Date: 20260508
Application Date: 20251125

Claims (7)

1. A pain assessment system for neonatal facial expression image analysis, comprising: the data acquisition module comprises an image acquisition unit, an image preprocessing unit and a feature point positioning unit, wherein the image acquisition unit is used for acquiring a neonatal facial expression image sequence, the image preprocessing unit is connected with the image acquisition unit and used for carrying out illumination compensation, noise removal and scale standardization on the facial expression image sequence, and the feature point positioning unit is connected with the image preprocessing unit and used for positioning facial key feature points including eyebrow points, eye corner points, nose point points and mouth corner points in the facial expression image sequence to obtain facial feature point coordinates; The facial expression manifold construction module is connected with the data acquisition module and comprises a characteristic point mapping unit, a Riemann metric construction unit and an individualization adjustment unit, wherein the characteristic point mapping unit is used for mapping the facial characteristic point coordinates to a low-dimensional Riemann manifold space through a nonlinear dimension reduction method, the Riemann metric construction unit is connected with the characteristic point mapping unit and is used for defining an adaptive Riemann metric tensor in the Riemann manifold space, wherein the weight coefficient of the adaptive Riemann metric tensor on the eyebrow, the periocular and the nasolabial regions is larger than the weight coefficient of the Riemann metric tensor on other facial regions, and the individualization adjustment unit is connected with the Riemann metric construction unit and is used for adjusting manifold parameters based on reference data in a pain-free state and constructing an individualization facial expression manifold model; The curvature field analysis module is connected with the facial expression manifold construction module and is used for calculating Gaussian curvature and average curvature for the individualized facial expression manifold model, generating a facial curvature field, tracking the change of the facial curvature field along with time to obtain dynamic characteristics of the curvature field, identifying curvature modes of eyebrow, eye periphery and nasal and lip sulcus areas and extracting micro-action curvature characteristics related to pain; The space-time characteristic analytic hierarchy process module is connected with the curvature field analytic module and is used for carrying out multi-scale decomposition on the micro-motion curvature characteristics in different time windows to form micro-scale, mesoscale and macro-scale characteristic representations, calculating space differential invariants, time differential invariants and space-time mixing invariants of curvature fields under each scale, carrying out self-adaptive weighted fusion on differential invariants of different scales and dimensions, and identifying pain related micro-motion units comprising eyebrow tightening lock, eye extrusion and nose-lip groove deepening based on the fused characteristics to generate micro-motion unit space-time characteristics; And the pain evaluation and alarm module is connected with the space-time characteristic hierarchical analysis module and is used for calculating the occurrence frequency, duration and intensity of the micro-action unit according to the space-time characteristics of the micro-action unit, generating a pain score by carrying out weighted summation on the occurrence frequency, duration and intensity, and executing alarm operation according to the pain score.
2. The pain assessment system of claim 1, wherein the curvature field analysis module comprises: A curvature calculation unit configured to calculate the gaussian curvature and the average curvature for the personalized facial expression manifold model, generating the facial curvature field; the curvature field evolution unit is connected with the curvature calculation unit and is used for tracking the change of the curvature field of the face along with time and acquiring the dynamic characteristics of the curvature field; And the micro-motion curvature rate characteristic extraction unit is connected with the curvature field evolution unit and is used for identifying specific curvature modes of the eyebrow, the periocular region and the nasolabial sulcus region and extracting the micro-motion curvature rate characteristic related to pain.
3. The pain assessment system for neonatal facial expression image analysis as set forth in claim 1, wherein the pain assessment and alarm module includes: A pain quantification unit for calculating the frequency, duration and intensity of occurrence of the micro-action unit, generating the pain score; a trend analysis unit connected with the pain quantification unit and used for tracking the time change trend of the pain score; a threshold adjustment unit, connected to the trend analysis unit, for dynamically adjusting the personalized pain threshold based on the historical data; And the alarm execution unit is connected with the threshold value adjustment unit and is used for triggering an alarm prompt of a corresponding level when the pain score exceeds the personalized pain threshold value.
4. The pain assessment system of claim 1, wherein the facial expression manifold construction module and the curvature field analysis module are connected by a feature data transmission interface, the feature data transmission interface is used for transmitting facial expression manifold data in real time and supporting synchronous analysis processing.
5. The pain assessment system of claim 1, wherein the temporal-spatial feature analysis module analyzes micro-deformations generated by instantaneous muscle contraction in a 0.1-0.5 second time window, analyzes a complete micro-motion unit expression process in a 0.5-2 second time window, and analyzes a combination pattern of a plurality of micro-motion units in a 2-10 second time window when performing multi-scale decomposition on the micro-motion rate features.
6. The pain assessment system for neonatal facial expression image analysis as set forth in claim 2, wherein the micro-motion curvature feature extraction unit is further configured to: identifying extreme points and contour lines in the face curvature field, and corresponding to the most obvious facial deformation area; Calculating a curvature gradient vector field, and describing the direction of the fastest change of the curvature of the face; a library of micro-curvature characteristics is created, storing typical curvature characteristic patterns associated with pain.
7. The pain assessment system for neonatal facial expression image analysis as set forth in claim 3, wherein the pain quantification unit calculates the pain score by a comprehensive scoring formula, wherein the comprehensive scoring formula sums the frequency, duration and intensity of occurrence of the micro-action unit with weights, the pain score being classified into four classes of mild, moderate, severe and critical for guiding clinical intervention.

Description

Pain assessment system for neonatal facial expression image analysis Technical Field The invention relates to the technical field of medical image processing, in particular to a pain assessment system for neonatal facial expression image analysis. Background The neonate cannot directly express pain feeling due to limited language expression ability, and medical staff often need to evaluate the pain condition of the neonate by observing indirect indexes such as facial expression, crying, limb actions and the like. At present, subjective assessment tools such as a neonatal facial expression pain score scale (NFCS), an infant pain score scale (NIPS) and the like are mainly used clinically, and have larger subjectivity and difference among scoring persons although the assessment tools have certain clinical applicability, so that objective, continuous and interference-free pain assessment is difficult to realize. With the development of computer vision and artificial intelligence technology, automated pain assessment methods based on facial expressions are receiving attention. In the prior art, pain classification is mainly performed by extracting facial feature points or texture features and combining a machine learning algorithm, but the facial expressions are generally regarded as static features in European space by the methods, manifold structures with facial expressions which are continuously deformed in nature are ignored, and fine changes and time dynamic characteristics of micro expressions are difficult to capture. Meanwhile, the conventional method generally adopts a universal model, and cannot effectively cope with individual differences of different newborns, so that the accuracy of an evaluation result is insufficient. Therefore, there is a need to develop a new-born facial expression image analysis system based on advanced geometric theory, which can accurately capture the micro-expression change related to pain, and realize objective, continuous and individual pain assessment. Disclosure of Invention The invention aims to provide a pain assessment system for analysis of neonatal facial expression images based on differential geometry theory, which realizes accurate assessment and timely early warning of neonatal pain states by constructing a facial expression manifold model, analyzing dynamic curvature field characteristics and a multi-scale space-time structure. The invention provides a pain assessment system for neonatal facial expression image analysis, which comprises: The data acquisition module is used for acquiring the facial expression images of the neonates and extracting facial feature point coordinates; the facial expression manifold construction module is connected with the data acquisition module and is used for receiving the facial feature point coordinates, mapping the facial feature point coordinates to a Riemann manifold space and constructing a facial expression manifold model; the curvature field analysis module is connected with the facial expression manifold construction module and is used for calculating a facial expression dynamic curvature field based on the facial expression manifold model and extracting curvature characteristics of a facial area; the space-time characteristic analytic hierarchy process module is connected with the curvature field analysis module and is used for carrying out multi-scale decomposition on the curvature characteristics, identifying the micro-motion units related to pain and generating space-time characteristics of the micro-motion units; And the pain evaluation and alarm module is connected with the space-time characteristic hierarchical analysis module and is used for receiving the space-time characteristics of the micro-action unit, calculating a pain score and executing alarm operation according to the pain score. Preferably, the data acquisition module includes: The image acquisition unit is used for acquiring a neonatal facial expression image sequence through the high-definition camera; the image preprocessing unit is connected with the image acquisition unit and is used for carrying out illumination compensation, noise removal and scale standardization on the facial expression image sequence; and the characteristic point positioning unit is connected with the image preprocessing unit and is used for positioning facial key characteristic points such as eyebrow points, eye corner points, nose tip points, mouth corner points and the like in the facial expression image sequence. Preferably, the facial expression manifold construction module includes: the feature point mapping unit is used for mapping the facial feature point coordinates to a low-dimensional Riemann manifold space through a nonlinear dimension reduction method; a Riemann metric construction unit connected with the feature point mapping unit for defining an adaptive Riemann metric tensor in the Riemann manifold space, wherein the adaptive Riemann metric tensor gives higher