Brigham Research Institute Poster Session Site logo-1
Close this search box.

Caroline Goedmakers, MD, PhD




Research Fellow




Research fellow




Caroline M.W. Goedmakers, Asad M. Lak, MD, Omar Arnaout, MD, Michael W. Groff, MD, Timothy R. Smith, MD, PhD, MPH, Carmen L.A. Vleggeert-Lankamp, M.Sc., MD, PhD, Hasan A. Zaidi, MD, Aakanksha Rana, PhD, M.Sc, Alessandro Boaro MD

Using Deep Learning to predict Adjacent Segment Disease on Pre-Operative Cervical MRI Images for Patients Undergoing Anterior Cervical Discectomy and Fusion

A vast amount of healthcare data is collected daily, can be harnessed to improve patientcare and bridge the gap between sick patients and healthy populations. I want to combine my future clinical job with my passion for machine learning analysis of these large population datasets. If not for my female scientific hero’s, my mentors, I had never uncovered this passion. However, women are still underrepresented in science and exactly that representation helped me learn I loved it. By participating I want to share my passion and hope to spark passion in someone who hasn’t uncovered it in herself, yet.


Patients who undergo surgery for cervical radiculopathy are at risk of developing adjacent segment disease (ASD) and in this study a deep learning algorithm was developed to  predict ASD using only the preoperative cervical MRI of patients undergoing single-level anterior cervical discectomy and fusion (ACDF).


Retrospective chart review was performed and after applying in- and exclusion criteria 344  patients were included of which 60% (n = 208) was used for training and 40% for validation (n = 43) and testing (n = 93). A deep learning-based prediction model was trained using preoperative T2-sagittal cervical MRI.


The model was able to predict ASD on the 93 test images with an accuracy of 88 / 93 [95%, CI 90.0 – 99.2], sensitivity 12 / 15 [80%, CI 59.8 – 100] and specificity 76 / 78 [97%, CI 93.9 – 100]. The model outperformed the neuroradiologist and neurosurgeon in all outcome metrics and McNemar’s test demonstrated that the proportion of wrongful predictions was significantly lower by the model (Test Statistic 2.000, p-value < 0.001).


A deep learning algorithm using only preoperative cervical T2 MRI outperformed a clinical expert at predicting new symptoms at adjacent levels.