Machine Learning Based Multi-Class Classification and Grading of Squamous Cell Carcinoma in Optical Microscopy.

Area of Science:

• Computational pathology
• Medical image analysis
• Oncology

Background:

• Histopathological tissue grading is crucial for disease diagnosis and treatment.
• Manual grading of squamous cell carcinoma (SCC) is time-consuming and requires expert pathologists.

Purpose of the Study:

• To develop and evaluate efficient machine learning (ML) and deep learning (DL) models for automated SCC histopathological image analysis.
• To compare the performance of various ML models and a CNN for SCC grading.

Main Methods:

• Extracted 360 features using Discrete Wavelet Transform, Gray Level Co-occurrence Matrix, and histogram analysis.
• Selected 114 key features using Student's t-test.
• Trained five ML models (SVM, Naïve Bayes, Decision Tree, KNN, Neural Network) with 5-, 7-, and 10-fold cross-validation.
• Trained a Convolutional Neural Network (CNN) for automated classification.

Main Results:

• The k-nearest neighbor (KNN) ML model with sevenfold cross-validation achieved 98% accuracy.
• The Convolutional Neural Network (CNN) model achieved 98.23% accuracy in automated SCC classification.
• Feature selection significantly improved model efficiency and accuracy.

Conclusions:

• Combining ML for feature analysis with interpretable DL models offers a pathway to accurate and efficient SCC grading.
• Automated analysis using ML and DL models can reduce reliance on manual histopathological assessments.
• This approach has the potential to improve diagnostic workflows in pathology.