🧬 Embryo Classification using Deep Learning

A deep learning project for classifying microscopic embryo images using a custom Convolutional Neural Network (CNN) built with TensorFlow and Keras.

The project explores the application of computer vision and deep learning to embryo image analysis, using a dataset containing two predefined image classes.

Project Type: Deep Learning / Computer Vision Domain: Healthcare AI • Medical Imaging

---

🎯 Overview

This project uses a CNN to learn visual patterns from microscopic embryo images and classify them into one of two dataset-defined classes.

The workflow covers image preprocessing, dataset preparation, CNN training, performance visualization, and single-image prediction.

Pipeline

Microscopic Embryo Image
          ↓
    Image Loading
          ↓
    Resize to 150×150
          ↓
     Dataset Shuffle
          ↓
     Train/Test Split
          ↓
       CNN Model
          ↓
   Feature Extraction
          ↓
    Softmax Classifier
          ↓
     Predicted Class

---

✨ Key Features

• 🔬 Microscopic embryo image classification
• 🧠 Custom CNN architecture using TensorFlow/Keras
• 🖼️ Image preprocessing with OpenCV
• ⚡ Batch normalization
• 🛡️ Dropout regularization
• 📊 Training and validation accuracy tracking
• 📉 Training and validation loss visualization
• 🔎 Single-image prediction
• 🔀 Dataset shuffling and train/test splitting
• 🏷️ One-hot encoded class labels

---

🧠 CNN Architecture

The model uses multiple convolutional layers to progressively learn visual features from the embryo images.

Input: 150 × 150 × 3
        │
        ▼
Conv2D (32)
        │
Conv2D (64)
        │
Batch Normalization
        │
MaxPooling
        │
Dropout
        │
        ▼
Conv2D (64)
        │
Conv2D (64)
        │
MaxPooling
        │
Dropout
        │
        ▼
Conv2D (128)
        │
Conv2D (128)
        │
Conv2D (128)
        │
MaxPooling
        │
Dropout
        │
        ▼
Conv2D (128)
        │
Conv2D (256)
        │
MaxPooling
        │
Dropout
        │
        ▼
Flatten
        │
Dense (512)
        │
Dense (512)
        │
Dropout
        │
        ▼
Dense (2)
        │
Softmax

Training Configuration

Parameter	Value
Input Size	150 × 150 × 3
Number of Classes	2
Optimizer	Adam
Loss Function	Categorical Cross-Entropy
Epochs	15
Validation Split	10%
Regularization	Dropout
Normalization	Batch Normalization
Framework	TensorFlow / Keras

---

🔬 Data Processing

Images are loaded from the dataset and resized to a fixed resolution before being passed to the model.

img = cv2.imread(image_path)
img = cv2.resize(img, (150, 150))

The dataset is then:

1. Loaded into NumPy arrays
2. Shuffled using a fixed random state
3. Split into training and test sets
4. Converted into categorical labels
5. Used to train the CNN

Label Encoding

The two dataset classes are represented as:

labels = ['0', '1']

The labels are converted into one-hot encoded vectors using TensorFlow.

---

📊 Training & Visualization

The model is trained for 15 epochs while tracking training and validation performance.

Training accuracy and validation accuracy are visualized using:

acc = history.history['accuracy']
val_acc = history.history['val_accuracy']

plt.plot(acc, label="Training Accuracy")
plt.plot(val_acc, label="Validation Accuracy")
plt.legend()
plt.show()

Training and validation loss are also plotted to observe the model's learning behavior.

loss = history.history['loss']
val_loss = history.history['val_loss']

plt.plot(loss, label="Training Loss")
plt.plot(val_loss, label="Validation Loss")
plt.legend()
plt.show()

---

🔎 Image Prediction

After training, the model can be used to classify an individual embryo image.

prediction = model.predict(img_array)
predicted_class = prediction.argmax()

The model returns probabilities for the two classes through the final softmax layer, and the class with the highest probability is selected as the prediction.

---

🛠️ Tech Stack

Machine Learning

• Python
• TensorFlow
• Keras
• Scikit-learn

Computer Vision

• OpenCV
• Pillow

Data Processing

• NumPy
• Pandas

Visualization

• Matplotlib
• Seaborn

Dataset

• Kaggle

---

⚠️ Limitations

This project is a deep learning prototype for image classification and is not a clinically validated medical diagnostic system.

• The model predicts the predefined classes present in the dataset.
• The dataset labels should not be interpreted as independent clinical diagnoses.
• Model performance depends on the quality and distribution of the training data.
• The current implementation uses 150 × 150 image resolution.
• The current evaluation focuses primarily on training and validation accuracy/loss.
• Predictions should not be used for medical decision-making.

---

💡 Key Takeaways

Through this project, I explored:

• Building CNN architectures from scratch
• Image preprocessing with OpenCV
• Preparing image datasets for deep learning
• Multi-class categorical classification
• Batch normalization and dropout
• Training and validation monitoring
• Model inference on individual images
• Applying computer vision techniques to healthcare-related data

---

👩‍💻 Author

Ayushi Taralkar

B.Tech Computer Science Engineering VIT Bhopal — Health Informatics

GitHub • LinkedIn