Enhanced Histopathologic Image Analysis for Mouth Cancer Classification Using Morphological Reconstruction and UNet.

Area of Science:

• Medical imaging and artificial intelligence
• Computational pathology
• Oncology diagnostics

Background:

• Mouth cancer is a significant global health issue with high mortality rates if not detected early.
• Traditional diagnostic methods for oral cancer are time-consuming, subjective, and prone to inter-observer variability.
• Existing deep learning models for cancer detection can be computationally intensive and less suitable for resource-constrained environments.

Purpose of the Study:

• To develop and evaluate a novel, efficient, and accurate deep learning framework, Depthwise Separable Convolution U-Net (DWSU-Net), for mouth cancer detection using histopathological images.
• To improve the accuracy and efficiency of oral cancer diagnosis by integrating advanced image preprocessing techniques with a lightweight convolutional neural network architecture.
• To enhance the interpretability of AI-driven diagnostic predictions for better clinical integration.

Main Methods:

• Development of the DWSU-Net architecture, integrating depthwise separable convolutions into a U-Net framework for reduced computational complexity.
• Implementation of a comprehensive preprocessing pipeline including Sobel filtering, Otsu thresholding, Canny edge detection, and morphological reconstruction via erosion.
• Training and evaluation of the DWSU-Net model on a publicly available Kaggle dataset of oral cancer histopathological images, using fivefold cross-validation.
• Comparative analysis with conventional Convolutional Neural Network (CNN) architectures to determine optimal model-filtering combinations.

Main Results:

• The DWSU-Net model, combined with morphological reconstruction preprocessing, achieved a classification accuracy of 99.74% in distinguishing between healthy and malignant oral tissue images.
• The DWSU-Net architecture demonstrated superior performance compared to conventional CNNs, attributed to its efficient depthwise separable convolutions and effective preprocessing pipeline.
• The morphological reconstruction preprocessing step enhanced the distinctness of malignant regions, improving interpretability for both the AI model and human pathologists.

Conclusions:

• The proposed DWSU-Net framework offers a highly accurate and efficient solution for automated mouth cancer detection from histopathological images.
• The integration of depthwise separable convolutions and morphological reconstruction provides a robust, scalable, and interpretable decision-support tool for clinical settings.
• DWSU-Net shows significant potential for clinical deployment, addressing the limitations of traditional diagnostic methods and improving early detection rates for oral cancer.