TY - JOUR
T1 - Differentiating spinal pathologies by deep learning approach
AU - Haim, Oz
AU - Agur, Ariel
AU - Gabay, Segev
AU - Azolai, Lee
AU - Shutan, Itay
AU - Chitayat, May
AU - Katirai, Michal
AU - Sadon, Sapir
AU - Artzi, Moran
AU - Lidar, Zvi
N1 - Publisher Copyright:
© 2023 Elsevier Inc.
PY - 2024/2
Y1 - 2024/2
N2 - BACKGROUND CONTEXT: Spinal pathologies are diverse in nature and, excluding trauma and degenerative diseases, includes infectious, neoplastic (either extradural or intradural), and inflammatory conditions. The preoperative diagnosis is made with clinical judgment incorporating lab findings and radiological studies. When the diagnosis is uncertain, a biopsy is almost always mandatory since the treatment is dictated by the type of pathology. This is an invasive, timely, and costly process. PURPOSE: The aim of this study was to develop a deep learning (DL) algorithm, based on preoperative MRI and post-operative pathological results, to differentiate between leading spinal pathologies. STUDY DESIGN: We retrospectively collected and analyzed clinical, radiological, and pathological data of patients who underwent spinal surgery or biopsy for various spinal pathologies between 2008 and 2022 at a tertiary center. The patients were stratified according to their pathological reports (the threshold for inclusion was set to 25 patients per diagnosis). METHODS: Preoperative MRI, clinical data, and pathological results were processed by a deep learning model built on the Fast.ai framework on top of the PyTorch environment. RESULTS: A total of 231 patients diagnosed with carcinoma (80), infection (57), meningioma (52), or schwannoma (42), were included in our model. The mean overall accuracy was 0.78±0.06 for the validation, and 0.93±0.03 for the test dataset. CONCLUSION: Deep learning algorithm for differentiation between the aforementioned spinal pathologies, based solely on clinical MRI, proves as a feasible primary diagnostic modality. Larger studies should be performed to validate and improve this algorithm for clinical use. CLINICAL SIGNIFICANCE: This study provides a proof-of-concept for predicting spinal pathologies solely by MRI based DL technology, allowing for a rapid, targeted, and cost-effective work-up and subsequent treatment.
AB - BACKGROUND CONTEXT: Spinal pathologies are diverse in nature and, excluding trauma and degenerative diseases, includes infectious, neoplastic (either extradural or intradural), and inflammatory conditions. The preoperative diagnosis is made with clinical judgment incorporating lab findings and radiological studies. When the diagnosis is uncertain, a biopsy is almost always mandatory since the treatment is dictated by the type of pathology. This is an invasive, timely, and costly process. PURPOSE: The aim of this study was to develop a deep learning (DL) algorithm, based on preoperative MRI and post-operative pathological results, to differentiate between leading spinal pathologies. STUDY DESIGN: We retrospectively collected and analyzed clinical, radiological, and pathological data of patients who underwent spinal surgery or biopsy for various spinal pathologies between 2008 and 2022 at a tertiary center. The patients were stratified according to their pathological reports (the threshold for inclusion was set to 25 patients per diagnosis). METHODS: Preoperative MRI, clinical data, and pathological results were processed by a deep learning model built on the Fast.ai framework on top of the PyTorch environment. RESULTS: A total of 231 patients diagnosed with carcinoma (80), infection (57), meningioma (52), or schwannoma (42), were included in our model. The mean overall accuracy was 0.78±0.06 for the validation, and 0.93±0.03 for the test dataset. CONCLUSION: Deep learning algorithm for differentiation between the aforementioned spinal pathologies, based solely on clinical MRI, proves as a feasible primary diagnostic modality. Larger studies should be performed to validate and improve this algorithm for clinical use. CLINICAL SIGNIFICANCE: This study provides a proof-of-concept for predicting spinal pathologies solely by MRI based DL technology, allowing for a rapid, targeted, and cost-effective work-up and subsequent treatment.
KW - Deep Learning
KW - Diagnosis
KW - MRI
KW - Machine learning
KW - Spine
UR - http://www.scopus.com/inward/record.url?scp=85175256422&partnerID=8YFLogxK
U2 - 10.1016/j.spinee.2023.09.019
DO - 10.1016/j.spinee.2023.09.019
M3 - ???researchoutput.researchoutputtypes.contributiontojournal.article???
C2 - 37797840
AN - SCOPUS:85175256422
SN - 1529-9430
VL - 24
SP - 297
EP - 303
JO - Spine Journal
JF - Spine Journal
IS - 2
ER -