Convolutional Neural Networks for Classifying Cervical Cancer Types Using Histological Images

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Cervical cancer is a leading global cancer in women.
• Accurate diagnosis and classification are crucial for effective treatment and survival.
• Histopathology is key for diagnosing cervical malignancies.

Purpose of the Study:

• To develop and validate a CNN-based system for cervical cancer diagnosis.
• To enhance diagnostic interpretability using AI visualization techniques.
• To compare CNN performance against human pathologists.

Main Methods:

• Utilized 8496 labeled histology images from 229 cervical specimens.
• Trained and validated CNN models (AlexNet, VGG-19, Xception, ResNet-50) using five-fold cross-validation.
• Employed Guided Backpropagation and Grad-CAM for interpretability.

Main Results:

• The Xception model demonstrated excellent performance in identifying cervical squamous cell carcinoma (SCC) and adenocarcinoma (AC).
• Achieved high AUC values for SCC (0.98 internal, 0.974 external) and AC (0.966 internal, 0.958 external).
• CNN performance was comparable to that of pathologists, with Grad-CAM highlighting malignant features.

Conclusions:

• CNNs are effective for classifying cervical malignancies in histological images.
• AI tools can aid pathologists by identifying specific areas of concern.
• CNNs show potential as a diagnostic aid in cervical cancer pathology.