Machine Learning and Deep Learning for the Diagnosis of Cervical Degenerative Diseases: Systematic Review and Meta-Analysis.
Hongwei Duan, Ruiyuan Chen, Minghui Liang et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis evaluated traditional machine learning (TML) and deep learning (DL) approaches for diagnosing cervical degenerative diseases, analyzing data from 30 studies involving 25,301 patients. The pooled sensitivity was 0.92 and specificity was 0.88, indicating strong diagnostic performance, though the overall certainty of evidence was rated as low.
Key findings
- Pooled sensitivity for TML and DL was 0.92 (95% CI 0.89-0.96).
- Pooled specificity was 0.88 (95% CI 0.84-0.91).
- Positive likelihood ratio was 8.36 (95% CI 6.14-11.36).
- Area under the SROC curve was 0.96 (95% CI 0.94-0.97).
Why it matters
The findings highlight the potential of AI technologies in enhancing diagnostic accuracy for cervical degenerative diseases, which is crucial given the rising incidence of these conditions. However, the low certainty of evidence suggests caution in their clinical application.
What to keep in mind
High risk of bias was identified in several studies, and external validation of the models was limited.