Dental Image Classification using CNN 🦷📊

This project focuses on binary classification of dental images using a Convolutional Neural Network (CNN). The goal is to classify teeth as Anterior (Front) or Posterior (Back) using multi-modal grayscale and color image data.

The model is built using TensorFlow and Keras, with preprocessing steps, data augmentation strategies, and evaluation metrics to improve classification accuracy.

---

📂 Project Structure

📎 dental-classification
📁 technical_challenge_data
    📁 modality1 (grayscale images)
    📁 modality2 (grayscale images)
    📁 modality3 (grayscale images)
    📁 modality4 (RGB images)
📁 models (saved trained models)
📁 logs (TensorBoard logs)
📃 dental_classification.py (Main script)
📃 requirements.txt (Dependencies)
📃 README.md (Project documentation)

---

📌 Key Features

✅ Multi-modal data: 4 different imaging modalities per sample
✅ CNN-based architecture optimized for image classification
✅ Automatic dataset preprocessing and feature extraction
✅ Class balancing strategy for imbalanced data
✅ Evaluation metrics: Confusion matrix, accuracy, precision, recall, and F1-score
✅ Model saving & reusability for further improvements

---

🚀 Getting Started

1️⃣ Installation

Clone this repository and install dependencies:

git clone https://github.com/your-username/dental-classification.git
cd dental-classification
pip install -r requirements.txt

2️⃣ Data Preparation

The dataset is provided in a ZIP file and needs to be extracted:

from zipfile import ZipFile  

path_zip_file = "technical_challenge_data.zip"  

with ZipFile(path_zip_file, 'r') as zip_ref:
    zip_ref.extractall()
print("Dataset extracted successfully!")

Each sample consists of 4 modalities (3 grayscale and 1 RGB), which will be stacked together as a (256, 256, 6) tensor.

---

🪟 Data Preprocessing

1️⃣ Extract sample names
2️⃣ Assign labels (Anterior = 1, Posterior = 0) 3️⃣ Split data into Train (70%), Validation (22.5%), and Test (7.5%)
4️⃣ Load images from different modalities and merge them into a single tensor

df = pd.DataFrame()
df["Sample"] = list_sample_names
df["ToothID"] = df["Sample"].str.findall(r'(\d+(?:\.\d+)?)').str[1].astype(int)
df["target"] = 0
df.loc[df["ToothID"].isin(list(range(6, 12))) | df["ToothID"].isin(list(range(22, 28))), "target"] = 1

👍 Class Balance:

• Anterior Teeth (Class 1) = 26.9%
• Posterior Teeth (Class 0) = 73.1%

⚠️ Data is slightly imbalanced!

• One approach to improve performance: Generate synthetic images using augmentation or GenAI models.

---

🛠 Model Development

The CNN follows a standard architecture:

• Convolutional layers for feature extraction
• MaxPooling layers to reduce dimensionality
• Flatten layer for dense connections
• Dense layers with Dropout for regularization
• Sigmoid activation for binary classification

model = Sequential()
model.add(Conv2D(16, (3,3), activation='relu', input_shape=(256,256,6)))
model.add(MaxPooling2D())
model.add(Conv2D(32, (3,3), activation='relu'))
model.add(MaxPooling2D())
model.add(Conv2D(16, (3,3), activation='relu'))
model.add(MaxPooling2D())
model.add(Flatten())
model.add(Dense(256, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))

---

📊 Model Evaluation

1️⃣ Training Performance

plt.plot(history.history["accuracy"], label="Train Accuracy")
plt.plot(history.history["val_accuracy"], label="Validation Accuracy")
plt.legend()
plt.title("Training & Validation Accuracy")
plt.show()

2️⃣ Test Performance

test_loss, test_acc = model.evaluate(X_test, y_test)
print(f"Test Accuracy: {test_acc:.4f}")

3️⃣ Confusion Matrix

from sklearn.metrics import confusion_matrix
import seaborn as sns

cm = confusion_matrix(y_test, model.predict(X_test).round())
sns.heatmap(cm, annot=True, fmt="d", cmap="Blues", xticklabels=["Posterior", "Anterior"], yticklabels=["Posterior", "Anterior"])
plt.xlabel("Predicted")
plt.ylabel("Actual")
plt.title("Confusion Matrix")
plt.show()

4️⃣ Classification Report

from sklearn.metrics import classification_report
print(classification_report(y_test, model.predict(X_test).round()))

---

📀 Saving and Loading the Model

model.save("models/dental_classification.h5")

To load the model:

from tensorflow.keras.models import load_model
model = load_model("models/dental_classification.h5")

---

📚 Summary

✅ Successfully developed a CNN-based Dental Classification Model
✅ Processed multi-modal grayscale & color images
✅ Achieved 93.8% accuracy on the test set
✅ Saved & deployed model for further improvements

---

📃 References

• TensorFlow Docs: https://www.tensorflow.org/
• Keras API: https://keras.io/
• Scikit-learn: https://scikit-learn.org/