ivf

AI-ML-Project-Embryo-Analysis-to-Improve-Success-Rate-of-IVF

This project is a recreation of the STORK repository Github Link to STORK used for embryo classification.

---

📖 Table of Contents

<details open="open"><summary>Table of Contents</summary>

1. About the Project
2. Prerequisites
3. Dataset
4. Data Pipeline
5. Methodology
6. Results and Discussion
7. References
8. Contributors
9. Steps To Follow

</details>

---

✏️ About The Project

Infertility is a clinical condition characterized by an inability to conceive after one year or longer of unprotected sex. In vitro fertilization (IVF) is a type of assistive reproductive technology (ART) for infertility treatment and surrogacy.

This project aims to use machine learning to classify embryos as "good" or "poor" based on their quality, improving IVF success rates and reducing human error in embryo selection.

<p align="center"> <img width="100%" src="https://www.pfcla.com/hubfs/infographic-1.svg"> </p>

---

🍴 Prerequisites

This project requires Python and Conda for setting up the environment.

Installation Steps:

1. Clone the repository:

   git clone https://huggingface.co/saikAIML/ivf
   cd ivf
   

2. Install Conda (if not already installed).
   Follow the instructions at Miniconda Installation.

3. Create and activate the Conda environment:

   conda create -n embryo-env python=3.7
   conda init
   conda activate embryo-env
   

4. Install required dependencies:

   pip install -r requirements.txt
   

---

▶️ Steps to Follow

Training the Model (Optional):

The model is already trained on Inception-V1 with the dataset and the weights are available at scripts/slim/result this is if you want use a different model

1. Download pre-trained Inception-V1 models from TensorFlow Pretrained Models.
2. Create the process directory :

   python convert.py ../Images/train process/ 0
   

3. Modify load_inception_v1.sh to set correct paths.
4. Run the script:

   ./run/load_inception_v1.sh
   

   OR use Python:

   python train_image_classifier.py --train_dir=scripts/result    --dataset_name=embryo --dataset_split_name=train    --dataset_dir=scripts/process --model_name=inception_v1    --checkpoint_path=run/checkpoint/inception_v1.ckpt    --checkpoint_exclude_scopes=InceptionV1/Logits    --max_number_of_steps=5000 --batch_size=32 --learning_rate=0.01    --save_interval_secs=100 --save_summaries_secs=100    --log_every_n_steps=300 --optimizer=rmsprop --weight_decay=0.00004    --clone_on_cpu=True
   

Testing the Model:

1. Navigate to the slim directory:

   cd scripts/slim
   

2. Run the test script:

   python tst.py
   

3. To test different images, modify tst.py:

   • Open tst.py in a text editor.
   • Locate the line with predict_image_label.
   • Change the file path to a different embryo image.

   Example:

   from pred import predict_image_label
   print("**************", predict_image_label("result", "../../Images/test/good_23765483_-15_3AA.jpg", "v1"), "**************")
   

   • Save the file and rerun:

     python tst.py
     

📂 Dataset

• Source: STORK Framework
• Training Images: 42 Good, 42 Poor
• Test Images: 14
• Total Images: 98 (JPG format)

<p align="center"> <img width="70%" src="Docs/dataset.jpg"> </p>

---

🎯 Data Pipeline

<p align="center"> <img width="100%" src="Docs/pipeline.png"> </p>

---

📜 Methodology

We use a Deep Neural Network (DNN) for embryo image analysis based on Google’s Inception-V1 architecture. The pre-trained STORK framework helps classify embryos into good or poor categories.

The dataset is divided as follows:

• 85% (84 images) for training
• 15% (14 images) for testing

---

🔍 Results and Discussion

• The model achieved 100% accuracy on the test set (14 images).
• Using this model could reduce IVF cycles and costs by selecting the best embryo.
• A web interface allows users to upload an embryo image and check its quality.

---

📚 References

1. Muhammad et al. (2020) - NIH Article
2. Kirillova et al. (2020) - "Should we transfer poor quality embryos?"
3. PFCLA - IVF Overview
4. STORK GitHub - STORK Framework

---

✍️ Contributors

👩‍🎓 Anjana Padikkal Veetil
📧 AP202@myscc.ca | GitHub

👩‍🎓 Nobin Ann Mathew
📧 NM91@myscc.ca | GitHub

👨‍🎓 Santosh Kumar Kantimahanti Lakshmi
📧 SK602@myscc.ca | GitHub

👨‍🎓 Amal Mathew
📧 AM252@myscc.ca | GitHub

🎯 Final Notes:

• The model is pre-trained, so only testing is required unless you want to retrain it.
• Modify tst.py to test with different images.

💡 This project was developed as part of the Data Analytics for Business (May 2021) course at St. Clair College.