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

A Convolutional Neural Network for Automated Detection of Cervical Ossification of the Posterior Longitudinal Ligament using Magnetic Resonance Imaging

STUDY DESIGN

Retrospective cohort study.

OBJECTIVE

We aimed to develop and validate a convolutional neural network (CNN) model to distinguish between cervical ossification of posterior longitudinal ligament (OPLL) and multilevel degenerative spinal stenosis using Magnetic Resonance Imaging (MRI) and to compare the diagnostic ability with spine surgeons.

SUMMARY OF BACKGROUND DATA

Some artificial intelligence models have been applied in spinal image analysis and many of promising results were obtained; however, there was still no study attempted to develop a deep learning model in detecting cervical OPLL using MRI images.

MATERIALS AND METHODS

In this retrospective study, 272 cervical OPLL and 412 degenerative patients underwent surgical treatment were enrolled and divided into the training (513 cases) and test dataset (171 cases). CNN models applying ResNet architecture with 34, 50, and 101 layers of residual blocks were constructed and trained with the sagittal MRI images from the training dataset. To evaluate the performance of CNN, the receiver operating characteristic curves of 3 ResNet models were plotted and the area under the curve were calculated on the test dataset. The accuracy, sensitivity, and specificity of the diagnosis by the CNN were calculated and compared with 3 senior spine surgeons.

RESULTS

The diagnostic accuracies of our ResNet34, ResNet50, and ResNet101 models were 92.98%, 95.32%, and 97.66%, respectively; the area under the curve of receiver operating characteristic curves of these models were 0.914, 0.942, and 0.971, respectively. The accuracies and specificities of ResNet50 and ResNet101 models were significantly higher than all spine surgeons; for the sensitivity, ResNet101 model achieved better values than that of the 2 surgeons.

CONCLUSION

The performance of our ResNet model in differentiating cervical OPLL from degenerative spinal stenosis using MRI is promising, better results were achieved with more layers of residual blocks applied.

MRI-based lesion quality score assessing ossification of the posterior longitudinal ligament of the cervical spine

BACKGROUND CONTEXT

No method currently exists for MRI-based determination of ossification of the posterior longitudinal ligament (OPLL) of the cervical spine using objective criteria.

PURPOSE

The purpose of this study was to develop an MRI-based score to determine whether a lesion represents a cervical OPLL lesion and to establish the objective diagnostic value.

STUDY DESIGN

Retrospective cohort in a single medical institution.

PATIENT SAMPLE

Thirty-five patients undergoing surgery for OPLL (Group A) and 99 patients undergoing cervical disc arthroplasty for soft disc herniation (Group B) between 2011 and 2020 were retrospectively included. All OPLL lesions on unenhanced MRI scan were correlated with a corresponding CT scan. Demographics were comparable between the two groups.

OUTCOME MEASURES (PHYSIOLOGIC MEASURES)

Using unenhanced magnetic resonance imaging (MRI), the T1- and T2- lesion quality (LQ) scores were calculated. Receiver operating characteristic (ROC) analysis was performed to calculate the area-under-the-curve (AUC) of both LQ scores as a predictor of the presence of OPLL. Computed tomography- (CT-) based Hounsfield unit (HU) values of OPLL lesions were obtained and compared with both LQ scores. The LQ scores for MRI scanners from different manufacturers were compared using Student's t test to confirm the validity of the LQ score by scanner type.

METHODS

The regions of interest for signal intensity (SI) were defined as the darkest site of the lesion and the cerebrospinal fluid (CSF) at the cerebellomedullary cistern. The T1 and T2 LQ scores were measured as the ratio of the SI at the darkest site of the lesion divided by the SI of the CSF.

RESULTS

The T1 and T2 LQ scores in Group A were significantly lower than those in Group B (p < 0.001). ROC analysis determined that T1 and T2 LQ scores of 0.46 and 0.07, respectively, could distinguish the presence of OPLL with an accuracy of 0.93 and 0.89, respectively (p<.001). When the T1 LQ score of the lesion is <0.46, a diagnosis of OPLL may be suspected with 100% sensitivity and 92.3% specificity. The HU of the lesion had a moderate negative correlation with the T1 LQ score (r=-0.665, p<.0001). Both LQ scores were unaffected by manufacturer type.

CONCLUSIONS

This study found a correlation between the MRI-based T1 LQ scores and CT-based HU value for identifying OPLL lesions. Additional studies will be needed to validate that the T1 LQ score from the unenhanced MRI scan can identify cervical OPLL.

Assessment of Fusion After Anterior Cervical Discectomy and Fusion Using Convolutional Neural Network Algorithm

BACKGROUND

A convolutional neural network (CNN) is a deep learning (DL) model specialized for image processing, analysis, and classification.

OBJECTIVE

In this study, we evaluated whether a CNN model using lateral cervical spine radiographs as input data can help assess fusion after anterior cervical discectomy and fusion (ACDF).

STUDY DESIGN

Diagnostic imaging study using DL.

PATIENT SAMPLE

We included 187 patients who underwent ACDF and fusion assessment with postoperative one-year computed tomography and neutral and dynamic lateral cervical spine radiographs.

OUTCOME MEASURES

The performance of the CNN-based DL algorithm was evaluated in terms of accuracy and area under the curve.

MATERIALS AND METHODS

Fusion or nonunion was confirmed by cervical spine computed tomography. Among the 187 patients, 69.5% (130 patients) were randomly selected as the training set, and the remaining 30.5% (57 patients) were assigned to the validation set to evaluate model performance. Radiographs of the cervical spine were used as input images to develop a CNN-based DL algorithm. The CNN algorithm used three radiographs (neutral, flexion, and extension) per patient and showed the diagnostic results as fusion (0) or nonunion (1) for each radiograph. By combining the results of the three radiographs, the final decision for a patient was determined to be fusion (fusion ≥2) or nonunion (fusion ≤1). By combining the results of the three radiographs, the final decision for a patient was determined as fusion (fusion ≥2) or nonunion (nonunion ≤1).

RESULTS

The CNN-based DL model demonstrated an accuracy of 89.5% and an area under the curve of 0.889 (95% confidence interval, 0.793-0.984).

CONCLUSION

The CNN algorithm for fusion assessment after ACDF trained using lateral cervical radiographs showed a relatively high diagnostic accuracy of 89.5% and is expected to be a useful aid in detecting pseudarthrosis.

Development of artificial intelligence for automated measurement of cervical lordosis on lateral radiographs

Cervical sagittal alignment is an essential parameter for the evaluation of spine disorders. Manual measurement is time-consuming and burdensome to measurers. Artificial intelligence (AI) in the form of convolutional neural networks has begun to be used to measure x-rays. This study aimed to develop AI for automated measurement of lordosis on lateral cervical x-rays. We included 4546 cervical x-rays from 1674 patients. For all x-rays, the caudal endplates of C2 and C7 were labeled based on consensus among well-experienced spine surgeons, the data for which were used as ground truth. This ground truth was split into training data and test data, and the AI model learned the training data. The absolute error of the AI measurements relative to the ground truth for 4546 x-rays was determined by fivefold cross-validation. Additionally, the absolute error of AI measurements was compared with the error of other 2 surgeons’ measurements on 415 radiographs of 168 randomly selected patients. In fivefold cross-validation, the absolute error of the AI model was 3.3° in the average and 2.2° in the median. For comparison of other surgeons, the mean absolute error for measurement of 168 patients was 3.1° ± 3.4° for the AI model, 3.9° ± 3.4° for Surgeon 1, and 3.8° ± 4.7° for Surgeon 2. The AI model had a significantly smaller error than Surgeon 1 and Surgeon 2 (P = 0.002 and 0.036). This algorithm is available at (https://ykszk.github.io/c2c7demo/). The AI model measured cervical spine alignment with better accuracy than surgeons. AI can assist in routine medical care and can be helpful in research that measures large numbers of images. However, because of the large errors in rare cases such as highly deformed ones, AI may, in principle, be limited to assisting humans.