Deepath-SCC: a deep learning model for accurate tissue origin identification in squamous cell carcinoma.

Area of Science:

• Digital pathology
• Artificial intelligence in oncology
• Computational pathology

Background:

• Squamous cell carcinoma (SCC) diagnosis is challenging due to overlapping features across organs.
• Determining the tissue of origin for SCC is crucial for effective treatment but difficult with conventional methods.

Purpose of the Study:

• To develop and validate Deepath-SCC, a deep learning model for pan-squamous cell carcinoma tissue origin identification.
• To assess the model's accuracy using hematoxylin and eosin-stained whole-slide images.

Main Methods:

• A retrospective cohort of 4217 whole slide images from various SCC types (nasopharyngeal, head and neck/esophageal, lung, cervical, urothelial) was used for training and validation.
• Deep learning model (Deepath-SCC) applied to digitized whole-slide images.
• Internal and external test sets were utilized to evaluate model performance.

Main Results:

• Deepath-SCC achieved 91.2% accuracy and 0.986 AUROC in the internal test set.
• High-confidence predictions (≥0.7914) improved overall accuracy to 96.2%.
• External validation demonstrated 86.1% accuracy and 0.972 AUROC.

Conclusions:

• Deep learning-based digital pathology shows feasibility for predicting SCC tissue of origin.
• Deepath-SCC offers an efficient, cost-effective computational tool to support diagnostic workflows.
• The model can be particularly beneficial in challenging or resource-limited settings.