Automated cutaneous squamous cell carcinoma grading using deep learning with transfer learning.

Area of Science:

• Dermatopathology
• Computational pathology
• Artificial intelligence in medicine

Background:

• Histological grading of cutaneous squamous cell carcinoma (cSCC) is essential for patient prognosis and treatment planning.
• Manual grading methods are often subjective and inefficient, leading to potential diagnostic variability.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for automated cSCC grading.
• To enhance diagnostic accuracy (ACC) and efficiency in cSCC classification.

Main Methods:

• Three deep neural network (DNN) architectures (AlexNet, GoogLeNet, ResNet-18) were trained using transfer learning on 300 cSCC histopathological images.
• Model performance was assessed using ACC, sensitivity (SN), specificity (SP), and area under the curve (AUC).
• Clinical validation involved comparing DNN predictions with pathologist diagnoses on 60 additional images.

Main Results:

• The DL models demonstrated high performance, with ACC exceeding 85%, SN over 85%, SP above 92%, and AUC greater than 97%.
• Strong agreement was observed between DNN predictions and pathologist diagnoses, as well as across different network architectures.
• The developed DL models are publicly accessible for further research and clinical application.

Conclusions:

• Deep learning shows significant promise for improving the objectivity and efficiency of cSCC diagnosis.
• Automated grading using DL has the potential to enhance diagnostic accuracy and ultimately improve patient outcomes in cSCC management.