🧠 Laryngeal Cancer Detection and Classification Using Deep Learning on Histopathological Images

Laryngeal cancer presents complex and often subtle symptoms, making early detection a significant challenge in medical diagnostics. This project leverages the power of deep learning and computer vision to classify histopathological tissue images into different stages of laryngeal cancer.

📌 Overview

Our research focuses on developing a robust and high-performing image classification model to aid in the early detection and classification of laryngeal cancer. Using deep learning architectures, we trained and tested multiple models on a curated and augmented dataset of histopathological images.

🧬 Dataset

• Source: Provided by an Italian researcher.
• Total Images: 1,320 original histopathological images.
• After Augmentation: 5,280 images.
• Classes:
  • He: Healthy
  • Hbv: Hypertrophic Blood Vessels
  • IPCL: IPCL-like Vessel
  • Le: Leukoplakia

The dataset was preprocessed and augmented (using techniques such as rotation, flipping, and contrast enhancement) to improve generalization and prevent overfitting.

🧠 Models Evaluated

We tested multiple deep-learning-based image classification models, including:

• ResNet50 ✅ (Best Performing)
• DenseNet121
• DenseNet201
• VGG16
• VGG19
• MobileNetV2
• InceptionV3

Each model was fine-tuned using transfer learning and evaluated on accuracy, precision, recall, and F1-score metrics.

🏆 Best Model

• Model: Fine-tuned ResNet50
• Accuracy: 99.62%
• This model outperformed all others and surpassed previous benchmark results in similar studies.
• Emphasis was also placed on precision and recall, which are critical in disease classification tasks.

📊 Evaluation Metrics

We focused on multiple performance indicators:

• Accuracy
• Precision
• Recall
• F1 Score

These metrics ensure the reliability and sensitivity of the model, especially for early-stage disease detection where false negatives must be minimized.

🔧 Technologies Used

• Python
• TensorFlow / Keras
• OpenCV
• NumPy / Pandas
• Scikit-learn
• Matplotlib / Seaborn

📂 Project Structure

├── data/
│   ├── raw/                  # Original images
│   └── augmented/            # Augmented dataset
├── models/                   # Saved model weights
├── notebooks/                # Jupyter notebooks for training & evaluation
├── utils/                    # Helper functions (e.g., preprocessing, augmentation)
├── results/                  # Plots and evaluation reports
└── main.py                   # Training pipeline

🚀 Future Work

• Deploy the model in a clinical decision support tool.
• Incorporate segmentation to localize cancerous regions.
• Expand the dataset with multi-source histopathological images.

🙌 Acknowledgments

Special thanks to the Italian researcher who provided the dataset used in this study. This work was inspired by the need for better tools in early-stage cancer detection and diagnostic support.