Evaluation of an enhanced ResNet-18 classification model for rapid On-site diagnosis in respiratory cytology
2025

Improved AI Model for Lung Cancer Diagnosis

Sample size: 739 publication 10 minutes Evidence: moderate

Author Information

Author(s): Gong Wei, Vaishnani Deep K., Jin Xuan-Chen, Zeng Jing, Chen Wei, Huang Huixia, Zhou Yu-Qing, Hla Khaing Wut Yi, Geng Chen, Ma Jun

Primary Institution: Lishui Municipal Central Hospital

Hypothesis

Can an enhanced ResNet-18 model improve the accuracy of on-site diagnosis in respiratory cytology?

Conclusion

The deep learning model shows promise as an aid for on-site diagnosis of respiratory cytology samples, but human expertise remains essential.

Supporting Evidence

  • The AI model's diagnostic accuracy was comparable to that of human experts.
  • AI assistance improved diagnostic efficiency for all combinations of AI and human cytopathologist.
  • Human diagnostic accuracy was influenced by years of experience.

Takeaway

This study shows that a computer program can help doctors quickly and accurately identify lung cancer from cell samples, but doctors still need to be involved.

Methodology

The study used 739 respiratory specimens and tested an AI model against human cytopathologists to evaluate diagnostic accuracy.

Potential Biases

Potential sampling bias due to lesion location and operator experience.

Limitations

The study was limited to images from one hospital, and the use of Diff-Quik staining may have affected diagnostic accuracy.

Participant Demographics

Cytopathologists with varying levels of experience were involved in the study.

Digital Object Identifier (DOI)

10.1186/s12885-024-13402-3

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication