Construction and validation of a deep learning prediction model for cervical instability

Abstract

BACKGROUND: Early prediction of cervical instability is crucial for the prevention and treatment of cervical spondylosis, and deep learning technology can provide robust support for intelligent prediction of cervical instability. OBJECTIVE: To develop a deep learning model of cervical instability based on cervical magnetic resonance imaging for early intelligent prediction of cervical instability. METHODS: This study recruited young and middle-aged participants (18-45 years), including both cervical instability patients and healthy controls, through the Spine Department Outpatient Clinic of Wangjing Hospital, China Academy of Chinese Medical Sciences, as well as community-based recruitment. All participants underwent cervical magnetic resonance imaging examinations. On the axial magnetic resonance imaging images, five key anatomical structures were manually annotated: intervertebral disc, facet, prevertebral muscle, deep muscle group in the back of the neck, and superficial muscle group in the back of the neck. A deep learning algorithm was then employed to develop a predictive model for cervical instability, utilizing both the original images and the delineated regions of interest. Finally, the model's predictive performance was systematically evaluated and validated. RESULTS AND CONCLUSION: (1) The study included a total of 308 young and middle-aged participants, comprising 196 individuals with cervical instability and 112 healthy controls. Based on enrollment time, the subjects' data were allocated to either the model training set or the test set. (2) The model demonstrated high predictive performance, with an area under the curve values of 0.97, an F1-score of 0.98, a precision of 0.98, and a recall of 0.97 in the training set. In the test set, these values were 0.97, 0.95, 1.00, and 0.90, respectively. (3) The results indicate that the deep learning model based on cervical magnetic resonance images can achieve early intelligent prediction of cervical instability with high predictive performance.

1. Introduction

Cervical instability is an early pathological stage of cervical spondylosis, affecting patients' physical and mental health, while disrupting the original mechanical balance of the cervical spine and accelerating its degenerative changes [1-2]. In the information age, the incidence of cervical instability among young and middle-aged individuals is continuously rising. Early identification and effective intervention of cervical instability are of significant clinical value in delaying the progression of cervical degenerative diseases. The stability of the cervical spine is maintained by two major systems: the dynamic balance system (dominated by muscle tissue) and the static balance system (mainly including vertebrae, intervertebral discs, and facet joints). The synergistic action of both ensures normal physiological activity and mechanical balance of the cervical spine [3-4]. Cervical instability is a concentrated manifestation of the disruption of the dynamic and static balance systems. In young and middle-aged individuals, cervical instability often begins with dynamic imbalance caused by cervical muscle strain, followed by static imbalance [5-6]. Based on this mechanism, the authors hypothesize that if abnormal indicators can be detected and quantified during the stage of dynamic imbalance, and the degree of imbalance evaluated, it would be possible to achieve early prediction of cervical instability.

In previous research, the author's team innovatively used machine learning algorithms to construct a radiomics model of cervical instability based on magnetic resonance imaging, initially achieving early intelligent prediction of cervical instability [7]. However, radiomics technology heavily relies on manual feature engineering, with complex processing and computational steps, and employs traditional machine learning algorithms whose expressive power is limited by shallow architectures, affecting the model's clinical application and promotion [8-9]. Therefore, this study introduces deep learning algorithms to build an end-to-end deep learning model for cervical instability based on cervical MRI images, aiming to improve prediction accuracy and clinical applicability, and to advance the field of early diagnosis of degenerative spinal diseases.