Encyclopedia
Please note this is an old version of this entry, which may differ significantly from the current revision.
Subjects:
Orthopedics
From enhancing imaging techniques to predicting patient outcomes, AI and ML are revolutionizing the way we approach spinal diseases. AI and ML have significantly improved spinal imaging by augmenting detection and classification capabilities, thereby boosting diagnostic accuracy. Predictive models have also been developed to guide treatment plans and foresee patient outcomes, driving a shift towards more personalized care.
- artificial intelligence
- machine learning
- predictive model
1. Introduction
The management of spinal diseases is on the cusp of a transformative shift precipitated by the emergence and integration of artificial intelligence (AI) and machine learning (ML) into the realm of standard medical care [1,2,3,4,5,6]. Rather than being a vision of the distant future, this shift towards an intelligence-based spinal care model is well underway, promising a host of potential applications, including diagnosis, treatment, and the anticipation of adverse events [1,2,3,4,5,6].
The advent of AI and ML in healthcare is not an isolated phenomenon but rather the logical outcome of decades of accumulated scientific and technological progress within computational and healthcare disciplines. AI and ML have transcended mere theoretical promise; they are already delivering tangible results in the present day [1,2,3,4,5,6]. One of the most striking examples of their efficacy lies in the realm of spinal imaging [7]. Sophisticated algorithms augment the creation and interpretation of spinal images, thereby enriching the decision-making data available to clinicians [7]. It is plausible that future radiologists will collaborate seamlessly with these AI-driven systems to deliver more precise and personalized care.
2. AI and ML in Spinal Care
AI and ML have seen significant developments and implementations in recent years, particularly in the domain of healthcare (Table 2). Spinal care—a critical aspect of the healthcare system—has been no exception to this trend. Over time, these technologies have been utilized in various capacities in the sphere of spinal care, ranging from disease diagnosis to treatment and even the prediction of adverse events (Table 3) [8,12,13,14,15].
Table 2. Representative Machine-Learning Models for Healthcare Applications.
| Model | Description | Pros | Cons |
|---|---|---|---|
| Convolutional Neural Networks (CNNs) | CNNs are widely used for image-based tasks in healthcare, such as medical imaging analysis, including classification, segmentation, and detection. They leverage specialized layers to extract features from images and have achieved remarkable success in areas such as radiology and pathology. |
|
|
| Recurrent Neural Networks (RNNs) | RNNs are suitable for sequential data analysis and have been applied in various healthcare tasks. They can capture dependencies over time, making them valuable for tasks, such as time-series analysis, patient monitoring, and natural language processing in electronic health records (EHRs). |
|
|
| Neural Networks | Neural networks, including multi-layer perceptrons (MLPs), are versatile models used in healthcare. They are composed of interconnected layers of artificial neurons, enabling them to learn complex patterns in both structured and unstructured healthcare data. They have been applied to various tasks, including disease diagnosis, risk prediction, and patient outcome analysis. |
|
|
| Support Vector Machines (SVMs) | SVMs are a popular class of supervised learning algorithms used in healthcare. They are effective for classification tasks and have been applied in various areas, including disease diagnosis, risk prediction, and outcome analysis, by mapping data into high-dimensional feature spaces. |
|
|
| Random Forests | Random forests are an ensemble learning method that combines multiple decision trees to make predictions. They are versatile and have been used in disease diagnosis, prognosis, and feature selection by leveraging their ability to handle high-dimensional data and identify important features. |
|
|
| Deep Belief Networks (DBNs) | DBNs are generative models that employ unsupervised learning to learn hierarchical representations of data. They have shown promise in healthcare tasks, such as genetic analysis and medical imaging, and in clinical decision support systems by capturing complex patterns in large datasets. |
|
|
| Natural Language Processing (NLP) | NLP techniques are used to process and analyze human language data. They involve various tasks, such as sentiment analysis, text classification, named entity recognition, machine translation, and question-answering systems, enabling the understanding and extraction of information from textual data. |
|
|
| Decision Trees | Decision trees are simple yet powerful models used for classification and regression tasks. They partition data based on features to form a tree-like structure and make predictions. Decision trees are interpretable and can handle both categorical and numerical data. |
|
|
Table 3. Potential Applications of AI in Spinal Disease Care.
| Area of Spinal Disease Care | Description |
|---|---|
| Diagnosis and Detection | AI can assist in the automated analysis of medical imaging data, such as MRI or CT scans, for the detection and segmentation of spinal conditions, such as spinal stenosis. AI algorithms can aid in accurate and efficient diagnosis, providing valuable insights for healthcare professionals. |
| Treatment Planning | AI can support healthcare professionals in personalized treatment planning for spinal diseases. By analyzing patient data, including medical images, clinical records, and outcomes, AI algorithms can help determine the most appropriate treatment options and assist in surgical technique selection. |
| Surgical Guidance | AI can provide real-time guidance during spinal surgeries. By integrating pre-operative imaging data and intraoperative feedback, AI systems can help surgeons navigate complex spinal anatomies and make informed decisions, leading to improved surgical outcomes. |
| Predictive Modeling | AI can develop predictive models to assess disease progression and treatment outcomes for spinal diseases. These models can aid in prognosticating patient outcomes, optimizing treatment strategies, and facilitating shared decision making between healthcare providers and patients. |
| Rehabilitation Support | AI can assist in designing personalized rehabilitation programs for patients with spinal diseases. By analyzing patient data, including movement patterns and sensor data, AI algorithms can provide customized recommendations and monitoring during the rehabilitation process. |
| Remote Monitoring | AI-enabled remote monitoring systems can help track and monitor patients with spinal diseases outside of healthcare facilities. These systems can provide continuous monitoring, detect changes in symptoms or movement patterns, and alert healthcare providers for timely intervention. |
AI and ML have gained significant traction in spinal disease diagnosis, treatment recommendation, and patient outcome prediction. One notable case study is the work of Jujjavarapu et al., who used a deep-learning model to predict surgical outcomes in patients with lumbar disc herniation and lumbar spinal stenosis [16]. The study demonstrated that the AI model outperformed a benchmark model (logistic regression) in predicting early surgery, achieving an AUC of 0.725 compared to 0.597.
Another compelling case study was presented by Halicka et al., who developed an AI algorithm capable of predicting patient-reported outcomes following lumbar spine surgery [17]. The study aimed to develop and externally validate prediction models for spinal surgery outcomes using multivariate regression and random-forest approaches. The study included patients who underwent lumbar spine surgery for degenerative pathology. The models were evaluated based on changes in back and leg pain intensity and Core Outcome Measures Index (COMI) scores. The models demonstrated good calibration and explained variations in the validation data. The discrimination ability ranged from 0.62 to 0.72, indicating moderate predictive performance. The most important predictors included age, baseline scores, type of degenerative pathology, previous surgeries, smoking status, morbidity, and hospital-stay duration.
In addition to its diagnostic applications, AI has shown promise in predicting hospital stays following spine surgery. Shahrestani et al. conducted a study in which algorithms were trained using preoperative and perioperative variables from a dataset of patients with spondylolisthesis [18]. In the study conducted by Shahrestani et al., the researchers aimed to develop k-nearest-neighbors (KNN) classification algorithms to identify patients at a higher risk of extended hospital length of stay (LOS) following spinal surgery for spondylolisthesis. They analyzed the Quality Outcomes Database (QOD) spondylolisthesis dataset, including preoperative and perioperative variables. Out of 608 enrolled patients, 544 met the inclusion criteria. The KNN models exhibited impressive predictive performance. Model 1 achieved an overall accuracy of 98.1%, a sensitivity of 100%, a specificity of 84.6%, a positive predictive value (PPV) of 97.9%, and a negative predictive value (NPV) of 100%. Model 2 demonstrated an overall accuracy of 99.1%, a sensitivity of 100%, a specificity of 92.3%, a PPV of 99.0%, and an NPV of 100%. Receiver operating characteristic (ROC) curve analysis revealed an area under the curve (AUC) of 0.998 for both models.
These case studies serve as tangible examples of the benefits that AI and ML can bring to spinal care. They highlight that AI is not merely a futuristic concept but a current tool that is being utilized to enhance patient care. However, it is important to note that the application of these technologies is still in its early stages, and further research and clinical trials are needed to refine these tools and fully unlock their potential.
The introduction of AI and ML in spinal care signifies a paradigm shift toward an AI-augmented care model. An understanding of the evolution of computation in this context is crucial to appreciate the potential impact on diagnosis, treatment, and adverse-event prediction.
Decision-tree models have been used in predicting hospital readmission, prolonged hospital say, surgical complication, and direct cost following surgery for spinal stenosis with a high degree of accuracy [19,20,21,22]. They have also been utilized for texture analysis of spinal stenosis from MR imaging.
Natural language processing (NLP)—another application of AI—has also been explored in the context of spinal care [25,26]. For instance, an NLP system was developed and found to have a higher sensitivity for identifying standard reporting characteristics for low back pain on radiologic imaging compared to its rule-based counterpart [25]. This suggests that future NLP systems using ML could potentially enhance pathology-specific word choice to refine diagnosis and treatment strategies.
In addition to the more traditional ML techniques, support vector machines (SVMs) have also been used to classify patients with low back pain based on progression following rehabilitation [27]. A model developed by Jiang et al. achieved a striking 100% sensitivity and a 93.75% accuracy, hinting at the possibility of preoperative identification of patients who may require additional or more intensive rehabilitation efforts [27].
Comparatively, ANNs have been tested against gold-standard diagnostic categorization of low back pain in patients. The results revealed a high sensitivity and specificity of 95.7% and 100%, respectively. The successful combination of these ML algorithms with additional diagnostic tests could potentially revolutionize the clinician’s diagnostic process [28].
3. AI and ML in Spinal Imaging
Spinal imaging is a cornerstone of the diagnosis and management of various spinal disorders. Traditionally, these images have been analyzed manually by clinicians—a process that can be time-consuming and susceptible to human error. However, the advent of AI and ML has transformed this situation, enabling automated, fast, and precise image interpretation.
One of the fundamental applications of AI in spinal care revolves around localization—a concept associated with object detection and classification [32,33,34,35,36]. This concept enables the identification and labeling of an object in an image and can be invaluable in detecting anomalies in the spine. The SVM represents one such ML model that has demonstrated effectiveness in this field. SVMs have been used in detecting incidental lumbar spine fractures on X-rays, predicting forces applied to the lower back during weighted loading, and even characterizing type 1 Gaucher disease based on bone microarchitecture [37,38].
Similarly, the random-forest model, another ML technique, has been employed to identify osteoporosis more effectively than traditional bone turnover markers alone. It has also been used in the screening of patients undergoing non-osteoporosis dedicated CT imaging for potential osteoporosis [39].
Neural networks have found use in predicting fractures, both of the spine and the hip. Notably, these networks have been trained to detect posterior-element spinal fractures in trauma patients using CT images. Further, they have been used to identify hip fractures using a combination of radiographs, patient traits, and hospital process variables. In an exciting development, researchers found that image recognition algorithms using both imaging and non-imaging data may primarily use non-imaging data [40].
Convoluted neural networks (CNNs)—an offshoot of neural networks—have been developed to characterize and classify alignment-related pathologies, such as kyphosis and scoliosis [41,42]. One such CNN generated by Jamaludin et al. from dual-energy X-ray absorptiometry (DEXA) scans was able to automate spine curve identification, boasting a sensitivity of 86.5%, a specificity of 96.9%, and an AUC (area under the ROC curve) of 0.80 [43]. This capability opens the possibility for earlier detection of alignment-related pathologies, such as scoliosis and kyphosis.
Regression techniques have also been incorporated in ML for spinal care, with logistic regression models predicting the development of neuromuscular scoliosis in pediatric patients with cerebral palsy. Linear regression models have been used for postoperative height gain following the correction of idiopathic scoliosis [44].
Clustering methods have been used to identify distinct subgroups within adolescent idiopathic scoliosis populations. However, it has been challenging to identify discriminatory characteristics for patient clustering in certain study sets [42].
AI has also been beneficial in diagnosing various types of spinal pathologies, including lumbar neural foraminal stenosis and central spinal stenosis [10,15,44,45]. Deep neural networks have been employed to automatically localize and grade multiple spinal regions. These ML methods have the potential to reduce the qualitative MRI grading time in large epidemiological studies. Similarly, deep neural networks have been utilized to automatically localize and grade multiple spinal regions to diagnose conditions such as lumbar neural foraminal stenosis [44,45,46].
In another study, Roller et al. applied CNNs to MRI images to predict the operative level of patients undergoing disc decompression surgery [36]. An algorithm has also been developed to predict patients at risk for re-herniation after microdiscectomy, achieving a recall of 0.80 and an accuracy of 0.70 [36].
AI has even been used to predict early-onset adjacent segment degeneration following anterior cervical discectomy fusion (ACDF) using an SVM on tabular data [47].
Overall, AI and ML techniques in spinal imaging have shown promising results in improving the accuracy, speed, and predictive capability of the diagnosis and treatment of spinal conditions. These developments hold great promise for improving patient outcomes and transforming the way spinal care is delivered. However, it is essential to continue to refine these AI and ML models, incorporating new insights and additional patient features to ensure their continued evolution and relevance in clinical practice.
4. Role of AI and ML in Spinal Rehabilitation
AI and ML have diverse applications in rehabilitation, including applications in the field of physical medicine and rehabilitation (PM&R). In rehabilitation, ML is utilized in various areas, including symbiotic neuroprosthetics, myoelectric control, brain–computer interfaces, perioperative medicine, musculoskeletal medicine, diagnostic imaging, patient data measurement, and clinical decision support [48,49,50]. AI has even been used to assess rehabilitative exercises based on machine indications [48,49,50,51]. Brain–computer interfaces (BCIs) have emerged as a novel approach in neurorehabilitation. By recording and decoding brain signals, BCIs aim to enhance motor imagery-based training, facilitate task execution through functional electrical stimulation or robotic orthoses, and understand cerebral reorganizations after injury. BCIs show potential in promoting recovery and can be adapted to a diverse population. However, controlled clinical trials are needed to validate their effectiveness in pathological conditions and compare them to traditional methods [48,49].
5. Ensuring Accuracy in AI-Driven Diagnosis and Treatment
AI and ML models are as good as the data they are trained on. Thus, the quality and diversity of the data used play a significant role in the validity of the AI model. If the training data are not representative of the broader population or the specific patient groups, the AI model may perform poorly when deployed in real-world settings. Therefore, using high-quality, diverse, and representative datasets during model training is essential for ensuring the validity of AI models [53].
The reliability of an AI system refers to its ability to consistently produce the same results under the same conditions. This is particularly important in healthcare settings where reliable predictions are crucial for clinical decision making. Variability in AI system performance can lead to different diagnoses or treatment plans for the same patient, which can have serious implications for patient care [54]. Ensuring the validity and reliability of AI systems in spinal care also involves external validation, where the performance of AI models is assessed using data that were not involved in the model’s training or initial validation. This is a crucial step to gauge the generalizability of AI models and their readiness for real-world clinical deployment [55].
Lastly, a system for continuous monitoring and evaluation should be in place. This allows for the detection of any changes in the performance of AI models over time, providing an opportunity to make necessary adjustments and updates [56].
In conclusion, the integration of AI and ML into spinal care promises many benefits, but it is incumbent upon us to ensure the accuracy and reliability of these AI-driven systems.
This entry is adapted from the peer-reviewed paper 10.3390/jcm12134188
This entry is offline, you can click here to edit this entry!
