NASSi Best Paper Interview
Dr. AliAsghar MohammadiNasrabadi, PhD
Department of Systems Design Engineering University of Waterloo Waterloo, ON, Canada
This year’s NASS International Annual Meeting in Taipei marked a milestone: for the first time ever, a Best Paper was highlighted to showcase exceptional research from the global spine community. The distinction went to “Automated Lenke classification for preoperative spine surgery by extracting anatomical landmarks from X-ray images using a deep learning approach” by AliAsghar MohammadiNasrabadi, PhD; Gemah Moammer, FRCSC; and John McPhee, PhD. In the following conversation, Dr. MohammadiNasrabadi shares insights into the study and its implications for the future of spine care.
1. What question is your research attempting to answer?
Our research addresses the question, can artificial intelligence (AI) be used to automatically and accurately perform spinopelvic assessment and Lenke classification for adolescent scoliosis from standard X-ray images? We aimed to simplify and standardize this complex diagnostic process by eliminating manual measurement variability and reducing the time burden on clinicians.
2. Please summarize your key findings and comment on the clinical significance.
We developed a physics-informed deep learning model that automatically detects anatomical landmarks and extracts spinopelvic parameters and landmarks—including Cobb angle, SS, PT, PI, LL, SVA, femur center, sacrum end plate, iliac crest, L1–L5, T12–T1, C7–C2, apex, Cobb angle, LSRS, TSM, and CSRS—from lateral and AP spine X-rays. Our method achieves surgeon-level accuracy with intraclass correlation coefficients (ICCs) > 0.9 and parameter prediction accuracies exceeding 90%.
Clinically, this tool enables fast, consistent, and objective assessments that support preoperative planning and longitudinal monitoring of spinal deformities with high confidence and reliability.
3. What is surprising/exciting/different about your research results?
Most exciting finding is the demonstration that deep learning can replicate and even exceed intra-surgeon consistency in measuring critical spinal parameters—without requiring specialized hardware or manual tools. The system not only supports automated landmark detection but also performs full Lenke classification, significantly aiding triage and surgical planning.
This tool is also valuable in training environments, helping residents and fellows learn accurate measurement techniques while improving overall workflow efficiency.
4. How can this research ultimately apply to or benefit spine patients?
Patients with scoliosis typically require frequent follow-up imaging and assessments. Our model ensures consistent measurement across time points, regardless of who interprets the X-ray. This consistency reduces diagnostic uncertainty, supports earlier intervention decisions, and improves surgical planning.
Additionally, automation shortens assessment times, reduces costs, and minimizes the chance of variability-induced misjudgment in treatment decisions.
5. In what ways do you envision your research influencing or shaping future directions in spine-related research, clinical practice, or health policy?
We hope that this work will provide a new standard for AI-assisted preoperative planning in spine care. As deep learning tools gain clinical acceptance, they can be integrated into PACS systems and digital health records to provide real-time, standardized assessments. This work may also influence health policy by demonstrating the cost-effectiveness of AI-assisted diagnostics in reducing inter- and intra-observer variability and enhancing care delivery in both urban and rural settings.
6. Is there anything else you would like readers to know about this paper?
While our study focused on adolescent scoliosis and Lenke classification, the framework can be extended to other spinal deformities such as kyphosis, hyperlordosis, and flat-back syndrome. Our method is robust across various image qualities and systems (conventional radiography, EOS, CT scans), supporting broader clinical deployment.
7. Do you have any photos or images that could help a reporter understand this research and explain it to their audience?
Yes, we’ve provided two comparative images that highlight the core contribution of our work. The first image shows the traditional manual approach, where surgeons annotate anatomical landmarks and manually compute spinopelvic parameters—a process that is time-intensive and subject to human error. The second image demonstrates the output of our automated AI model, which detects landmarks and extracts parameters from both AP and lateral X-ray views. These features are then used to automatically classify patients based on the Lenke system. Together, these visualizations clearly illustrate the transition from manual to AI-driven assessment and how our system enhances efficiency, consistency, and accuracy in spinal deformity evaluation.
Figure 1. Manual Approach: In this image, the traditional manual workflow is depicted, where spine surgeons must individually annotate anatomical landmarks on X-ray images to compute spinopelvic parameters. These values are then manually recorded and referenced during surgery. The process is not only time-consuming but also susceptible to human error and inter-observer variability, impacting consistency and efficiency in clinical planning.
Figure 2. Automated Approach: This figure illustrates the output of our AI model, showcasing automatically detected anatomical landmarks and computed spinopelvic parameters on both AP and lateral X-ray images. These extracted features enable automated Lenke classification, streamlining the assessment process and enhancing accuracy, consistency, and efficiency in preoperative planning.
8. Would you like to add anything else?
I would like to express my sincere gratitude to Professor John McPhee, Canada Research Chair in Biomechatronics at the University of Waterloo, for his outstanding mentorship and continued guidance throughout this project. I am also deeply thankful to Dr. Gemah Moammer, Head of Spine Surgery at Waterloo Regional Health Network (WRHN), for his invaluable clinical expertise and collaboration.
Additionally, I would like to thank Andre Hladio, CTO and Co-Founder of Intellijoint Surgical, for his support and encouragement during various stages of this work.