Download This Paper
Preprints with The Lancet is a collaboration between The Lancet Group of journals and SSRN to facilitate the open sharing of preprints for early engagement, community comment, and collaboration. Preprints available here are not Lancet publications or necessarily under review with a Lancet journal. These preprints are early-stage research papers that have not been peer-reviewed. The usual SSRN checks and a Lancet-specific check for appropriateness and transparency have been applied. The findings should not be used for clinical or public health decision-making or presented without highlighting these facts. For more information, please see the FAQs.
Deep Learning Performance in Detecting Spine Metastases and Evaluating Features from CT Imaging: A Multi-Center Test Study
21 Pages Posted: 13 Mar 2023
More...Abstract
Background: Spine metastases could lead to pathologic fractures, which cause a sharp decrease in the quality of life and increase the mortality. The deep learning system (DLS) can be valuable for improving doctors’ performances. This study aimed to assess the effect of assistance by deep learning system (DLS) on diagnostic performances of doctors for spine metastases detection and features evaluation from CT imaging.
Methods: This prospective study used the multi-reader, multi-case methodology based on an external multicenter dataset between December 2021 and June 2022. Eighteen doctors were presented the test dataset with and without DLS assistance, with a 1-month washout period. Sensitivity and specificity per lesion was calculated to compare the diagnostic accuracy for spine metastases detection and features evaluation with and without DLS assistance.
Findings: A total of 78 patients (181 metastatic vertebrae) were in the test dataset. The sensitivity and specificity was higher with DLS aid than without DLS (96.9% [95% CI: 96.4, 97.4] vs 95.5% [95% CI: 94.6, 96.3] and 95.7% [95% CI: 94.6, 96.7] vs 94.8% [95% CI: 93.6, 96.0]; P values < 0.001) for spine metastases detection, with resident doctors improved significantly higher than attending or chief doctors (P values < 0.05). With DLS assistance, the sensitivity and specificity increased significantly for bone lesion quality detection (P values < 0.001), for vertebral body collapse detection (73.2% [95% CI: 67.6, 78.8] vs 83.6% [95% CI: 81.1, 86.1] and 85.4% [95% CI: 84.0, 86.7] vs 87.8% [95% CI: 86.7, 88.8]; P values < 0.001), and posterolateral involvement detection (70.5% [95% CI: 65.5, 75.5] vs 78.6% [95% CI: 75.5, 81.7] and 83.6% [95% CI: 81.8, 85.3] vs 87.5% [95% CI: 86.3, 88.7]; P values < 0.001), with resident doctors increased most sensitivities and specificities significantly higher than attending or chief doctors (P values < 0.05).
Interpretation: We successfully developed a DLS for automatic spine metastases detection and features evaluation from CT imaging. DLS assistance improved the performance of doctors, with the most significant benefit for resident doctors.
Funding: This study was supported by National Natural Science Foundation of China (81602519, 81672852), Shanghai Shen Kang three-year plan of action project (16CR2007A), Peak Plateau Project of Shanghai Jiao Tong University (20172024), and Interdisciplinary program of Shanghai Jiao Tong University (YG2017MS15).
Declaration of Interest: All authors declare there is no conflict of interest.
Ethical Approval: This study was approved by the ethics committee of Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine (2016-102). Principles of the Declaration of Helsinki were followed. Written informed consent was obtained from all participants in this study.
Keywords: deep learning system, spine metastases detection, features evaluation, CT
Suggested Citation: Suggested Citation