Artificial intelligence clustering of adult spinal deformity sagittal plane morphology predicts surgical characteristics, alignment, and outcomes

Thoracolumbar fusions for adult lumbar deformity show superior QALY gain and lower costs compared with upper thoracic fusions

PURPOSE

Adult spinal deformity (ASD) patients with sagittal plane deformity (N) or structural lumbar/thoraco-lumbar (TL) curves can be treated with fusions stopping at the TL junction or extending to the upper thoracic (UT) spine. This study evaluates the impact on cost/cumulative quality-adjusted life year (QALY) in patients treated with TL vs UT fusion.

METHODS

ASD patients with > 4-level fusion and 2-year follow-up were included. Index and total episode-of-care costs were estimated using average itemized direct costs obtained from hospital records. Cumulative QALY gained were calculated from preoperative to 2-year postoperative change in Short Form Six-Dimension (SF-6D) scores. The TL and UT groups comprised patients with upper instrumented vertebrae (UIV) at T9-T12 and T2-T5, respectively.

RESULTS

Of 566 patients with type N or L curves, mean age was 63.2 ± 12.1 years, 72% were female and 93% Caucasians. Patients in the TL group had better sagittal vertical axis (7.3 ± 6.9 vs. 9.2 ± 8.1 cm, p = 0.01), lower surgical invasiveness (- 30; p < 0.001), and shorter OR time (- 35 min; p = 0.01). Index and total costs were 20% lower in the TL than in the UT group (p < 0.001). Cost/QALY was 65% lower (492,174.6 vs. 963,391.4), and 2-year QALY gain was 40% higher, in the TL than UT group (0.15 vs. 0.10; p = 0.02). Multivariate model showed TL fusions had lower total cost (p = 0.001) and higher QALY gain (p = 0.03) than UT fusions.

CONCLUSION

In Schwab type N or L curves, TL fusions showed lower 2-year cost and improved QALY gain without increased reoperation rates or length of stay than UT fusions.

LEVEL OF EVIDENCE

III.

Artificial Intelligence Assistance for the Measurement of Full Alignment Parameters in Whole-Spine Lateral Radiographs

BACKGROUND

Measuring spinal alignment with radiological parameters is essential in patients with spinal conditions likely to be treated surgically. These evaluations are not usually included in the radiological report. As a result, spinal surgeons commonly perform the measurement, which is time-consuming and subject to errors. We aim to develop a fully automated artificial intelligence (AI) tool to assist in measuring alignment parameters in whole-spine lateral radiograph (WSL X-rays).

METHODS

We developed a tool called Vertebrai that automatically calculates the global spinal parameters (GSPs): Pelvic incidence, sacral slope, pelvic tilt, L1-L4 angle, L4-S1 lumbo-pelvic angle, T1 pelvic angle, sagittal vertical axis, cervical lordosis, C1-C2 lordosis, lumbar lordosis, mid-thoracic kyphosis, proximal thoracic kyphosis, global thoracic kyphosis, T1 slope, C2-C7 plummet, spino-sacral angle, C7 tilt, global tilt, spinopelvic tilt, and hip odontoid axis. We assessed human-AI interaction instead of AI performance alone. We compared the time to measure GSP and inter-rater agreement with and without AI assistance. Two institutional datasets were created with 2267 multilabel images for classification and 784 WSL X-rays with reference standard landmark labeled by spinal surgeons.

RESULTS

Vertebrai significantly reduced the measurement time comparing spine surgeons with AI assistance and the AI algorithm alone, without human intervention (3 minutes vs. 0.26 minutes; P < 0.05). Vertebrai achieved an average accuracy of 83% in detecting abnormal alignment values, with the sacral slope parameter exhibiting the lowest accuracy at 61.5% and spinopelvic tilt demonstrating the highest accuracy at 100%. Intraclass correlation analysis revealed a high level of correlation and consistency in the global alignment parameters.

CONCLUSIONS

Vertebrai's measurements can accurately detect alignment parameters, making it a promising tool for measuring GSP automatically.

Machine learning clustering of adult spinal deformity patients identifies four prognostic phenotypes: a multicenter prospective cohort analysis with single surgeon external validation

BACKGROUND CONTEXT

Among adult spinal deformity (ASD) patients, heterogeneity in patient pathology, surgical expectations, baseline impairments, and frailty complicates comparisons in clinical outcomes and research. This study aims to qualitatively segment ASD patients using machine learning-based clustering on a large, multicenter, prospectively gathered ASD cohort.

PURPOSE

To qualitatively segment adult spinal deformity patients using machine learning-based clustering on a large, multicenter, prospectively gathered cohort.

STUDY DESIGN/SETTING

Machine learning algorithm using patients from a prospective multicenter study and a validation cohort from a retrospective single center, single surgeon cohort with complete 2-year follow up.

PATIENT SAMPLE

About 805 ASD patients; 563 patients from a prospective multicenter study and 242 from a single center to be used as a validation cohort.

OUTCOME MEASURES

To validate and extend the Ames-ISSG/ESSG classification using machine learning-based clustering analysis on a large, complex, multicenter, prospectively gathered ASD cohort.

METHODS

We analyzed a training cohort of 563 ASD patients from a prospective multicenter study and a validation cohort of 242 ASD patients from a retrospective single center/surgeon cohort with complete two-year patient-reported outcomes (PROs) and clinical/radiographic follow-up. Using k-means clustering, a machine learning algorithm, we clustered patients based on baseline PROs, Edmonton frailty, age, surgical history, and overall health. Baseline differences in clusters identified using the training cohort were assessed using Chi-Squared and ANOVA with pairwise comparisons. To evaluate the classification system's ability to discern postoperative trajectories, a second machine learning algorithm assigned the single-center/surgeon patients to the same 4 clusters, and we compared the clusters' two-year PROs and clinical outcomes.

RESULTS

K-means clustering revealed four distinct phenotypes from the multicenter training cohort based on age, frailty, and mental health: Old/Frail/Content (OFC, 27.7%), Old/Frail/Distressed (OFD, 33.2%), Old/Resilient/Content (ORC, 27.2%), and Young/Resilient/Content (YRC, 11.9%). OFC and OFD clusters had the highest frailty scores (OFC: 3.76, OFD: 4.72) and a higher proportion of patients with prior thoracolumbar fusion (OFC: 47.4%, OFD: 49.2%). ORC and YRC clusters exhibited lower frailty scores and fewest patients with prior thoracolumbar procedures (ORC: 2.10, 36.6%; YRC: 0.84, 19.4%). OFC had 69.9% of patients with global sagittal deformity and the highest T1PA (29.0), while YRC had 70.2% exhibiting coronal deformity, the highest mean coronal Cobb Angle (54.0), and the lowest T1PA (11.9). OFD and ORC had similar alignment phenotypes with intermediate values for Coronal Cobb Angle (OFD: 33.7; ORC: 40.0) and T1PA (OFD: 24.9; ORC: 24.6) between OFC (worst sagittal alignment) and YRC (worst coronal alignment). In the single surgeon validation cohort, the OFC cluster experienced the greatest increase in SRS Function scores (1.34 points, 95%CI 1.01-1.67) compared to OFD (0.5 points, 95%CI 0.245-0.755), ORC (0.7 points, 95%CI 0.415-0.985), and YRC (0.24 points, 95%CI -0.024-0.504) clusters. OFD cluster patients improved the least over 2 years. Multivariable Cox regression analysis demonstrated that the OFD cohort had significantly worse reoperation outcomes compared to other clusters (HR: 3.303, 95%CI: 1.085-8.390).

CONCLUSION

Machine-learning clustering found four different ASD patient qualitative phenotypes, defined by their age, frailty, physical functioning, and mental health upon presentation, which primarily determines their ability to improve their PROs following surgery. This reaffirms that these qualitative measures must be assessed in addition to the radiographic variables when counseling ASD patients regarding their expected surgical outcomes.

Knee flexion compensation in postoperative adult spinal deformity patients: implications for sagittal balance and clinical outcomes

PURPOSE

To determine whether maintaining good sagittal balance with significant knee flexion (KF) constitutes a suboptimal outcome after adult spinal deformity (ASD) correction.

METHODS

This single-center, single-surgeon retrospective study, assessed ASD patients who underwent posterior spinal fusion between 2014 and 2020. Inclusion criteria included meeting at least one of the following: PI-LL ≥ 25°, T1PA ≥ 20°, or CrSVA-H ≥ 2 cm. Those with lower-extremity contractures were excluded. Patients were classified into four groups based on their 6-week postoperative cranio-hip balance and KF angle, and followed for at least 2 years: Malaligned with Knee Flexion (MKF+) (CrSVA-H > 20 mm + KFA > 10), Malaligned without Knee Flexion (MKF-) (CrSVA-H > 20 mm + KFA < 10), Aligned without Knee Flexion (AKF-) (CrSVA-H < 20 mm + KFA < 10), and Aligned with Knee Flexion (AKF+) (CrSVA-H < 20 mm + KFA > 10). The primary outcomes of this study included one and two year reoperation rates. Secondy outcomes included clinical and patient reported outcomes.

RESULTS

263 patients (mean age 60.0 ± 0.9 years, 74.5% female, and mean Edmonton Frailty Score 3.3 ± 0.2) were included. 60.8% (160/263 patients) exhibited good sagittal alignment at 6-week postop without KF. Significant differences were observed in 1-year (p = 0.0482) and 2-year reoperation rates (p = 0.0374) across sub-cohorts, with the lowest and highest rates in the AKF- cohort (5%, n = 8) and MKF + cohort (16.7%, n = 4), respectively. Multivariable Cox regression demonstrated the AKF- cohort exhibited significantly better reoperation outcomes compared to other groups: AKF + (HR: 5.24, p = 0.025), MKF + (HR: 31.7, p < 0.0001), and MKF- (HR: 11.8, p < 0.0001).

CONCLUSION

Our findings demonstrate that patients relying on knee flexion compensation in the early postoperative period have inferior outcomes compared to those achieving sagittal balance without knee flexion. When compared to malaligned patients, those with CrSVA-H < 20 mm and KFA > 10 degrees experience fewer early reoperations but similar delayed reoperation rates. This insight emphasizes the importance of considering knee compensation perioperatively when managing sagittal imbalance in clinical practice.

Artificial Intelligence to Preoperatively Predict Proximal Junction Kyphosis Following Adult Spinal Deformity Surgery: Soft Tissue Imaging May Be Necessary for Accurate Models

STUDY DESIGN

Retrospective cohort.

OBJECTIVE

In a cohort of patients undergoing adult spinal deformity (ASD) surgery, we used artificial intelligence to compare three models of preoperatively predicting radiographic proximal junction kyphosis (PJK) using: (1) traditional demographics and radiographic measurements, (2) raw preoperative scoliosis radiographs, and (3) raw preoperative thoracic magnetic resonance imaging (MRI).

SUMMARY OF BACKGROUND DATA

Despite many proposed risk factors, PJK following ASD surgery remains difficult to predict.

MATERIALS AND METHODS

A single-institution, retrospective cohort study was undertaken for patients undergoing ASD surgery from 2009 to 2021. PJK was defined as a sagittal Cobb angle of upper-instrumented vertebra (UIV) and UIV+2>10° and a postoperative change in UIV/UIV+2>10°. For model 1, a support vector machine was used to predict PJK within 2 years postoperatively using clinical and traditional sagittal/coronal radiographic variables and intended levels of instrumentation. Next, for model 2, a convolutional neural network (CNN) was trained on raw preoperative lateral and posterior-anterior scoliosis radiographs. Finally, for model 3, a CNN was trained on raw preoperative thoracic T1 MRIs.

RESULTS

A total of 191 patients underwent ASD surgery with at least 2-year follow-up and 89 (46.6%) developed radiographic PJK within 2 years. Model 1: Using clinical variables and traditional radiographic measurements, the model achieved a sensitivity: 57.2% and a specificity: 56.3%. Model 2: a CNN with raw scoliosis x-rays predicted PJK with a sensitivity: 68.2% and specificity: 58.3%. Model 3: a CNN with raw thoracic MRIs predicted PJK with average sensitivity: 73.1% and specificity: 79.5%. Finally, an attention map outlined the imaging features used by model 3 elucidated that soft tissue features predominated all true positive PJK predictions.

CONCLUSIONS

The use of raw MRIs in an artificial intelligence model improved the accuracy of PJK prediction compared with raw scoliosis radiographs and traditional clinical/radiographic measurements. The improved predictive accuracy using MRI may indicate that PJK is best predicted by soft tissue degeneration and muscle atrophy.

Alterations in Magnitude and Shape of Thoracic Kyphosis Following Surgical Correction for Adult Spinal Deformity

STUDY DESIGN

Retrospective review of prospective multicenter data.

OBJECTIVES

This study aimed to investigate the shape of TK before and after fusion in ASD patients treated with long fusion.

METHODS

ASD patients undergoing posterior spinal fusions including at least T5 to L1 without prior fusion extending to the thoracic spine were included. Patients were categorized based on the preoperative T1-T12 kyphosis into: Hypo-TK (if < 30°), Normal-TK, and Hyper-TK (if > 70°). Regional kyphosis at T10-L1 (Distal), T5-T10 (Middle), and T1-T5 (Proximal) and their relative contributions to total kyphosis were compared between groups, and the pre-to postoperative changes were investigated using paired t test.

RESULTS

In total, 329 patients were included in this analysis (mean age: 57 ± 16 years, 79.6% female). Preoperative T1-T12 TK for the entire cohort was 40.9 ± 2° (32% Hypo-TK, 11% Hyper-TK, 57% Normal-TK). The Hypo-TK group had the smallest distal TK (5.9 vs 17.1 & 26.0), and middle TK (8.0 vs 25.3 & 45.4), but the percentage of contribution to total kyphosis was not significantly different (Distal: 24.1% vs 34.1% vs 32.8%; Middle: 46.6% vs 53.9% vs 56.8%, all P > .1). Postoperatively, T1-12 TK increased significantly (40.9 ± 2.0° vs 57.8 ± 17.6°). Each group had a decrease in distal kyphosis (Hypo-TK 2.6 ± 10.4°; Normal-TK 8.9 ± 11.5°; Hyper-TK 14.9 ± 12°, all P < .05). The middle kyphosis significantly decreased for Hyper-TK (11.8 ± 12.4) and increased for both Normal-TK and Hypo-TK (3.8 ± 11° and 14.2 ± 11°). Proximal TK increased significantly for all groups by 14-18°. Deterioration from Normal-TK to Hyper-TK postoperatively was associated with lower rate of patient satisfaction (59.6% vs 77.3%, P = .032).

CONCLUSIONS

Posterior spinal fusion for ASD alters the magnitude and shape of thoracic kyphosis. While 60% of patients had a normal TK at baseline, 30% of those patients developed iatrogenic hyperkyphosis postoperatively. Patients with baseline hypokyphosis were more likely to be corrected to normal TK than hyperkyphotic patients. Future research should investigate TK restoration in ASD and its impact on clinical outcomes and complications.

Emerging Technologies within Spine Surgery

New innovations within spine surgery continue to propel the field forward. These technologies improve surgeons' understanding of their patients and allow them to optimize treatment planning both in the operating room and clinic. Additionally, changes in the implants and surgeon practice habits continue to evolve secondary to emerging biomaterials and device design. With ongoing advancements, patients can expect enhanced preoperative decision-making, improved patient outcomes, and better intraoperative execution. Additionally, these changes may decrease many of the most common complications following spine surgery in order to reduce morbidity, mortality, and the need for reoperation. This article reviews some of these technological advancements and how they are projected to impact the field. As the field continues to advance, it is vital that practitioners remain knowledgeable of these changes in order to provide the most effective treatment possible.

Machine Learning for the Orthopaedic Surgeon: Uses and Limitations

➤

Machine learning is a subset of artificial intelligence in which computer algorithms are trained to make classifications and predictions based on patterns in data. The utilization of these techniques is rapidly expanding in the field of orthopaedic research.

➤

There are several domains in which machine learning has application to orthopaedics, including radiographic diagnosis, gait analysis, implant identification, and patient outcome prediction.

➤

Several limitations prevent the widespread use of machine learning in the daily clinical environment. However, future work can overcome these issues and enable machine learning tools to be a useful adjunct for orthopaedic surgeons in their clinical decision-making.

The application of artificial intelligence in spine surgery

Due to its obvious advantages in processing big data and image information, the combination of artificial intelligence and medical care may profoundly change medical practice and promote the gradual transition from traditional clinical care to precision medicine mode. In this artical, we reviewed the relevant literatures and found that artificial intelligence was widely used in spine surgery. The application scenarios included etiology, diagnosis, treatment, postoperative prognosis and decision support systems of spinal diseases. The shift to artificial intelligence model in medicine constantly improved the level of doctors' diagnosis and treatment and the development of orthopedics.

An Artificial Neural Network Model for the Prediction of Perioperative Blood Transfusion in Adult Spinal Deformity Surgery

Prediction of blood transfusion after adult spinal deformity (ASD) surgery can identify at-risk patients and potentially reduce its utilization and the complications associated with it. The use of artificial neural networks (ANNs) offers the potential for high predictive capability. A total of 1173 patients who underwent surgery for ASD were identified in the 2017–2019 NSQIP databases. The data were split into 70% training and 30% testing cohorts. Eighteen patient and operative variables were used. The outcome variable was receiving RBC transfusion intraoperatively or within 72 h after surgery. The model was assessed by its sensitivity, positive predictive value, F1-score, accuracy (ACC), and area under the curve (AUROC). Average patient age was 56 years and 63% were female. Pelvic fixation was performed in 21.3% of patients and three-column osteotomies in 19.5% of cases. The transfusion rate was 50.0% (586/1173 patients). The best model showed an overall ACC of 81% and 77% on the training and testing data, respectively. On the testing data, the sensitivity was 80%, the positive predictive value 76%, and the F1-score was 78%. The AUROC was 0.84. ANNs may allow the identification of at-risk patients, potentially decrease the risk of transfusion via strategic planning, and improve resource allocation.

AI Prediction of Neuropathic Pain after Lumbar Disc Herniation—Machine Learning Reveals Influencing Factors

The treatment options for neuropathic pain caused by lumbar disc herniation have been debated controversially in the literature. Whether surgical or conservative therapy makes more sense in individual cases can hardly be answered. We have investigated whether a machine learning-based prediction of outcome, regarding neuropathic pain development, after lumbar disc herniation treatment is possible. The extensive datasets of 123 consecutive patients were used to predict the development of neuropathic pain, measured by a visual analogue scale (VAS) for leg pain and the Oswestry Disability Index (ODI), at 6 weeks, 6 months and 1 year after treatment of lumbar disc herniation in a machine learning approach. Using a decision tree regressor algorithm, a prediction quality within the limits of the minimum clinically important difference for the VAS and ODI value could be achieved. An analysis of the influencing factors of the algorithm reveals the important role of psychological factors as well as body weight and age with pre-existing conditions for an accurate prediction of neuropathic pain. The machine learning algorithm developed here can enable an assessment of the course of treatment after lumbar disc herniation. The early, comparative individual prediction of a therapy outcome is important to avoid unnecessary surgical therapies as well as insufficient conservative therapies and prevent the chronification of neuropathic pain.

An Evolution Gaining Momentum—The Growing Role of Artificial Intelligence in the Diagnosis and Treatment of Spinal Diseases

In recent years, applications using artificial intelligence have been gaining importance in the diagnosis and treatment of spinal diseases. In our review, we describe the basic features of artificial intelligence which are currently applied in the field of spine diagnosis and treatment, and we provide an orientation of the recent technical developments and their applications. Furthermore, we point out the possible limitations and challenges in dealing with such technological advances. Despite the momentary limitations in practical application, artificial intelligence is gaining ground in the field of spine treatment. As an applying physician, it is therefore necessary to engage with it in order to benefit from those advances in the interest of the patient and to prevent these applications being misused by non-medical partners.

Combined Artificial Intelligence Approaches Analyzing 1000 Conservative Patients with Back Pain-A Methodological Pathway to Predicting Treatment Efficacy and Diagnostic Groups

Patients with back pain are common and present a challenge in everyday medical practice due to the multitude of possible causes and the individual effects of treatments. Predicting causes and therapy efficien cy with the help of artificial intelligence could improve and simplify the treatment. In an exemplary collective of 1000 conservatively treated back pain patients, it was investigated whether the prediction of therapy efficiency and the underlying diagnosis is possible by combining different artificial intelligence approaches. For this purpose, supervised and unsupervised artificial intelligence methods were analyzed and a methodology for combining the predictions was developed. Supervised AI is suitable for predicting therapy efficiency at the borderline of minimal clinical difference. Non-supervised AI can show patterns in the dataset. We can show that the identification of the underlying diagnostic groups only becomes possible through a combination of different AI approaches and the baseline data. The presented methodology for the combined application of artificial intelligence algorithms shows a transferable path to establish correlations in heterogeneous data sets when individual AI approaches only provide weak results.