Using Deep Learning to Predict PD-L1 in Lung Cancer
Author Information
Author(s): Wang Qiushi, Deng Xixiang, Huang Pan, Ma Qiang, Zhao Lianhua, Feng Yangyang, Wang Yiying, Zhao Yuan, Chen Yan, Zhong Peng, He Peng, Ma Mingrui, Feng Peng, Xiao Hualiang
Primary Institution: Daping Hospital, Army Medical University, Chongqing, China
Hypothesis
Can deep learning models accurately predict PD-L1 expression in lung squamous cell carcinoma using H&E stained images?
Conclusion
The deep learning model effectively segments and predicts PD-L1 expression in lung squamous cell carcinoma, aiding in treatment guidance.
Supporting Evidence
- The model achieved a dice similarity coefficient of 80% and an intersection over union of 72%.
- The root mean square error of the model's predictions was 26.8.
- The intra-group correlation coefficient with the gold standard was 0.92.
- AI predictions had a lower root mean square error compared to pathologists.
Takeaway
Researchers used a computer program to help doctors see if a lung cancer sample has a specific marker that helps in treatment, making it easier and more accurate.
Methodology
A dataset of H&E-stained images was created, and a deep learning model called TransUnet was used to segment and predict PD-L1 expression.
Potential Biases
Subjective interpretation differences among pathologists could affect results.
Limitations
The model's training group was small, and it has not been tested on lung adenocarcinoma.
Participant Demographics
Surgical excision samples from lung squamous cell carcinoma patients.
Statistical Information
P-Value
0.92
Confidence Interval
95% CI: 0.90–0.93
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website