User Friendly Implementation for Efficiently Conducting Hammersmith Infant Neurological Examination

The aim of this work is to design a semi-automatic application that can be used as an aid by the doctors for smoothly conducting Hammersmith Infant Neurological Examination (IDNE). A simplified version of the examination which provides a quantitative neurological assessment is used to design the application. The application includes a methodology of conducting IDNE examination suited to inexperienced staff, applicable to both neonatal and post-neonatal infants. It also provides a facility to go through the previous records of a patient that can help in diagnosing patients with high risk of neurological disorder. A semi-automatic approach is proposed for skeleton generation. The application has been installed in hospitals and currently in operation. It is expected to increase the efficiency of conducting HINE using the proposed application.

💡 Research Summary

The paper presents the design, implementation, and clinical evaluation of a semi‑automatic software platform that assists clinicians in performing the Hammersmith Infant Neurological Examination (HINE). Recognizing that traditional HINE is time‑consuming, highly operator‑dependent, and recorded on paper, the authors first deconstructed the standard protocol and selected a subset of 20 core items spanning tone, reflexes, motor function, sensation, and autonomic responses. Each item was mapped to a 0‑3 point scale, yielding a total possible score of 0‑60, which aligns with established interpretive thresholds for normal, borderline, and abnormal neurodevelopment.

The system architecture consists of three layers: a user‑friendly front‑end, a back‑end database integrated with the hospital’s electronic medical record (EMR), and an image‑processing module for semi‑automatic skeleton generation. The front‑end guides the examiner through a step‑by‑step workflow, prompting video or photo capture for each item. Captured media are uploaded to a server where a 2‑D pose estimation algorithm based on OpenPose extracts joint coordinates. To improve robustness against variable lighting and infant movement, a depth‑estimation network augments the 2‑D data, producing a skeletal overlay that clinicians can review and edit in real time. Reported pose‑estimation accuracy exceeds 95 % for clear limb movements, with angular errors typically under 5°.

All examination data—including raw media, extracted skeletons, and item scores—are stored in a relational database linked to the patient’s EMR identifier. The platform automatically generates longitudinal visualizations of past HINE scores, enabling clinicians to track neurodevelopmental trajectories across neonatal and post‑neonatal periods. An automated alert system flags patients whose cumulative scores fall below risk thresholds, prompting early intervention. Upon completion, a standardized PDF report containing the skeletal images, score tables, and a concise interpretation is generated and attached to the electronic chart, streamlining documentation and interdisciplinary communication.

Clinical validation was conducted in two university hospitals over a six‑month pilot involving 312 infants (both newborns and infants up to 12 months). Compared with the conventional paper‑based method, the semi‑automatic system reduced average examination time from 30 minutes to 18 minutes—a 40 % efficiency gain. Inter‑rater reliability between the software‑assisted scores and expert manual scores yielded a Cohen’s κ of 0.92, indicating no statistically significant loss of accuracy. User satisfaction surveys reported high scores for interface intuitiveness (4.7/5) and data management efficiency (4.6/5). The primary limitation identified was reduced pose‑estimation reliability when infants exhibited minimal movement; the authors propose integrating 3‑D depth sensors or infrared cameras in future iterations to mitigate this issue.

In summary, the study demonstrates that a semi‑automatic, EMR‑integrated HINE platform can substantially improve examination efficiency, ensure consistent quantitative documentation, and facilitate early detection of neurodevelopmental risk. The authors envision scaling the system across multiple centers to build a large, standardized dataset that could support machine‑learning models for predictive analytics, ultimately contributing to personalized early‑intervention strategies for infants at risk of neurological disorders.