Abstract 1140P
Background
Artificial intelligence (AI) models have been widely shown to be useful in making pathological diagnoses, and we previously established a reliable AI model for distinguishing major lung cancer subtypes (adenocarcinoma [ADC], squamous cell carcinoma [SCC] and small cell lung cancer [SCLC]) with extremely high accuracy using samples obtained from transbronchial lung biopsy (TBLB). In the current study, we validated the accuracy of this AI model using a large cohort of samples obtained from TBLB.
Methods
In the previous study, we trained a convolution neural network based on the EfficientNet-B1 architecture to classify ADC, SCC, SCLC, and non-neoplastic lesions from TBLB specimen whole slide images (WSIs; 70, 23, 12 and 171 specimens of ADC, SCC, SCLC and non-neoplastic lesions, respectively) using a training dataset of 276 images. The WSIs were manually annotated by pathologists by drawing around the regions that contained each subtype. We used a transfer learning approach, in which the starting weights were obtained from a pre-trained model on ImageNet. The model was then trained on our dataset using a supervised learning approach. To classify a WSI, the model was applied in a sliding window fashion with an input tile size of 224x224 and a stride of 128 on a magnification of x10. The maximum probability was then used as a WSI diagnosis. In this study, this established model was validated in 2171 TBLB specimen WSIs (439, 143, 73 and 1516 specimens of ADC, SCC, SCLC and non-neoplastic lesions, respectively).
Results
The model achieved a Receiver Operator Curve Area Under the Curve (ROC-AUC) of 0.9458 (95% confidence interval [CI] 0.9316-0.9575) for non-neoplastic lesions with an accuracy, sensitivity and specificity of 0.8738, 0.8971 and 0.8661, respectively. The ROC-AUCs for ADC, SCC and SCLC were 0.8469 (95% CI 0.8170-0.8725), 0.9141 (95% CI 0.8825-0.9443) and 0.9412 (95% CI 0.905-0.9703), respectively.
Conclusions
Our model achieved high ROC-AUCs for the differentiation of ADC, SCC, SCLC and non-neoplastic lesions. Further studies are warranted to refine the established model.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.