Predicting the Evolution of Lung Squamous Cell Carcinoma In Situ Using Computational Pathology.

Area of Science:

• Oncology
• Computational Pathology
• Digital Pathology

Background:

• Lung squamous cell carcinoma in situ (SCIS) is a preinvasive lesion with unpredictable progression to invasive cancer.
• Approximately one-third of SCIS lesions regress spontaneously, posing a challenge for overtreatment.
• Predicting SCIS lesion evolution is crucial for effective patient management.

Purpose of the Study:

• To explore computational pathology for predicting the progression of SCIS to lung squamous cell carcinoma (SCC).
• To evaluate the performance of pathomics and deep learning models in forecasting SCIS evolution.

Main Methods:

• Utilized 112 H&E-stained whole slide images (WSIs) of SCIS lesions from the Image Data Resource.
• Trained a pathomics-based ridge classifier using 2000 features and a deep convolutional neural network (ResNet18).
• Assessed model performance using metrics like F1-score, precision, and recall.

Main Results:

• The pathomics model achieved an F1-score of 0.77, precision of 0.80, and recall of 0.77.
• The deep learning model demonstrated comparable performance with a WSI-level F1-score of 0.80, precision of 0.71, and recall of 0.90.
• Both computational pathology approaches showed potential in predicting SCIS progression.

Conclusions:

• Computational pathology offers valuable insights into the evolutionary behavior of SCIS.
• Larger datasets are necessary to further enhance model accuracy.
• Future applications may include predicting outcomes for other preinvasive lesions.