Abstract:Abstract Although the role of plain radiographs in diagnosing lumbar spinal stenosis (LSS) has declined in importance since the advent of magnetic resonance imaging (MRI), diagnostic ability of plain radiographs has improved dramatically when combined with deep learning. Previously, we developed a convolutional neural network (CNN) model using a radiograph for diagnosing LSS. In this study, we aimed to improve and generalize the performance of CNN models and overcome the limitation of the single-pose-based CNN (SP-CNN) model using multi-pose radiographs. Individuals with severe or no LSS, confirmed using MRI, were enrolled. Lateral radiographs of patients in three postures were collected. We developed a multi-pose-based CNN (MP-CNN) model using the encoders of the three SP-CNN model (extension, flexion, and neutral postures). We compared the validation results of the MP-CNN model using four algorithms pretrained with ImageNet. The MP-CNN model underwent additional internal and external validations to measure generalization performance. The ResNet50-based MP-CNN model achieved the largest area under the receiver operating characteristic curve (AUROC) of 91.4% (95% confidence interval [CI] 90.9–91.8%) for internal validation. The AUROC of the MP-CNN model were 91.3% (95% CI 90.7–91.9%) and 79.5% (95% CI 78.2–80.8%) for the extra-internal and external validation, respectively. The MP-CNN based heatmap offered a logical decision-making direction through optimized visualization. This model holds potential as a screening tool for LSS diagnosis, offering an explainable rationale for its prediction.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: by using multi - posture (multi - body position) radiographic images, develop a model based on convolutional neural network (CNN) to improve the diagnostic accuracy of lumbar spinal stenosis (LSS) and overcome the limitations of single - body position CNN model (SP - CNN). Specifically, the researchers found that although the importance of traditional planar radiographic images in the diagnosis of lumbar spinal stenosis has decreased after the emergence of magnetic resonance imaging (MRI), the diagnostic ability of planar radiographic images has been significantly improved after combining with deep - learning techniques. However, single - body position radiographic images have limitations in clinical diagnosis, because spine surgeons usually need to observe radiographic images from multiple perspectives and positions in order to diagnose diseases more accurately. Therefore, the researchers developed a multi - body position CNN model (MP - CNN), which can simultaneously process lateral radiographic images of patients in three different postures (extension, flexion and neutral postures) to improve the accuracy and generalization performance of diagnosis. ### Main objectives: 1. **Improve diagnostic accuracy**: By combining multi - body position radiographic images, improve the diagnostic accuracy of the CNN model for LSS. 2. **Enhance the generalization ability of the model**: Through internal and external validations, evaluate the performance of the model on different data sets to ensure its reliability in actual clinical applications. 3. **Provide interpretable diagnostic basis**: By generating heatmaps, show the areas of attention of the model in images of different postures, providing interpretable diagnostic basis for doctors. ### Methods: - **Data collection**: From May 1, 2005 to December 31, 2017, lateral radiographic images of patients with severe or no LSS confirmed by MRI were collected. - **Model construction**: A multi - body position CNN model (MP - CNN) was developed. This model uses three different encoders to process radiographic images in extension, flexion and neutral postures respectively. The feature maps extracted by these encoders are spliced together and then classified by multi - layer perceptron (MLP). - **Model validation**: Four different pre - training algorithms (VGG16, VGG19, ResNet50 and EfficientNet - B1) were used for internal and external validations to evaluate the performance of the model. ### Results: - **Performance indicators**: In internal validation, the AUROC of the ResNet50 - based MP - CNN model reached 91.4% (95% CI 90.9 - 91.8%). In additional internal validation and external validation, this model also performed well, with AUROC of 91.3% (95% CI 90.7 - 91.9%) and 79.5% (95% CI 78.2 - 80.8%) respectively. - **Heatmap analysis**: By generating heatmaps, the areas of attention of the model in images of different postures were shown, which is helpful for explaining the prediction results of the model. ### Conclusion: The multi - body position CNN model (MP - CNN) performs well in the diagnosis of lumbar spinal stenosis, which not only improves the diagnostic accuracy, but also enhances the generalization ability and interpretability of the model. This model is expected to become an effective tool for LSS screening.

Multi-pose-based convolutional neural network model for diagnosis of patients with central lumbar spinal stenosis

Lumbar Spinal Stenosis Grading in Multiple Level Magnetic Resonance Imaging Using Deep Convolutional Neural Networks

Deep Learning Model for Automated Detection and Classification of Central Canal, Lateral Recess, and Neural Foraminal Stenosis at Lumbar Spine MRI

Deep learning algorithm to evaluate cervical spondylotic myelopathy using lateral cervical spine radiograph

Feasibility of deep learning algorithm in diagnosing lumbar central canal stenosis using abdominal CT

Cervical Spondylosis Diagnosis Based on Convolutional Neural Network with X-ray Images

Detecting ossification of the posterior longitudinal ligament on plain radiographs using a deep convolutional neural network: a pilot study

Deep Learning Model for Automated Detection and Classification of Central Canal and Neural Foraminal Stenosis at Cervical Spine MRI

Convolutional neural network algorithm trained on lumbar spine radiographs to predict outcomes of transforaminal epidural steroid injection for lumbosacral radicular pain from spinal stenosis

AI-based lumbar central canal stenosis classification on sagittal MR images is comparable to experienced radiologists using axial images

Artificial Intelligence Detection of Cervical Spine Fractures Using Convolutional Neural Network Models

Radiograph-based rheumatoid arthritis diagnosis via convolutional neural network

Machine Learning Predicts Decompression Levels for Lumbar Spinal Stenosis Using Canal Radiomic Features from Computed Tomography Myelography

DeepSPINE: Automated Lumbar Vertebral Segmentation, Disc-level Designation, and Spinal Stenosis Grading Using Deep Learning

Part Affinity Fields and CoordConv for Detecting Landmarks of Lumbar Vertebrae and Sacrum in X-ray Images

Machine learning-based automated scan prescription of lumbar spine MRI acquisitions

Executing CNN-LSTM Algorithm for Recognizable Proof of Cervical Spondylosis Infection on Spinal Cord MRI Image

Classification of cervical spine disease using convolutional neural network

Deep learning model for automated diagnosis of degenerative cervical spondylosis and altered spinal cord signal on MRI

Convolutional Neural Network-Based Diagnostic Model for a Solid, Indeterminate Solitary Pulmonary Nodule or Mass on Computed Tomography

Automated Differentiation Between Osteoporotic Vertebral Fracture and Malignant Vertebral Fracture on MRI Using a Deep Convolutional Neural Network