This repository contains both the original research project and its extended interactive simulation version developed for the classification of thyroid nodules using advanced Machine Learning (ML), Deep Learning (DL), and a Hybrid modeling approach.
Originally submitted as a final year project for the Bachelor of Engineering in Computer Engineering at King Mongkut's University of Technology Thonburi (KMUTT), this work is now extended into a user-friendly online educational simulation tool to facilitate broader accessibility, especially for non-technical users, clinicians, and medical researchers.
- Ms. Suhani Mehta (64070503483)
- Ms. Krisha Botadara (64070503484)
- Dr. Jaturon Harnsomburana, Ph.D. (Advisor)
- Asst. Prof. Dr.-Ing Priyakorn Pusawiro, Ph.D. (Co-Advisor)
- Dr. Piyanit Wepulanon, Ph.D. (Committee Member)
- Dr. Aye Hninn Khine, Ph.D. (Committee Member)
This research investigates ML, DL, and hybrid modeling approaches for classifying thyroid nodules using ultrasound images. Two datasets were used: DDTI (images + clinical metadata) and TR12345 (TIRADS-labeled images from KMUTT). The models implemented include:
- ML on clinical metadata
- DL (custom CNNs)
- Hybrid DL feature extraction + ML classifier
The hybrid approach supports both binary (benign/malignant) and multiclass (TIRADS) classification.
Keywords: medical imaging, ultrasound, machine learning, deep learning, thyroid nodules, TIRADS, hybrid model
-
Evaluate ML, DL, and hybrid models for thyroid classification
-
Compare performance across models using TIRADS score
-
Design a custom CNN for binary classification (benign vs malignant)
-
Utilize pre-trained models for hybrid model
-
Test models on two distinct datasets (DDTI and TR12345)
-
Assess model generalizability and robustness
-
Provide insights for future AI applications in medical imaging
An extended simulation platform was built using Next.js, TailwindCSS, Framer Motion, and SVG-based animations to visually represent each stage of the pipeline. This simulation aims to make the research process interactive and understandable for a wider audience.
Landing page of the simulation platform
-
Dataset upload & visualization
-
Preprocessing pipeline: feature extraction, encoding, normalization, PCA
-
Hybrid model pipeline simulation (DL feature extraction + ML classification)
Hybrid feature extraction and classification pipeline
Feature preprocessing steps visualized in the simulation
Animated model training visualization for RF, DT, SVM, XGB -
Animated training pipeline for RF, DT, SVM, XGB
-
Binary & multiclass classification options
-
Result metrics visualization
-
User feedback integration via Google Forms
📌 Explore the simulation here: Interactive Platform
- Python, Scikit-learn, TensorFlow/Keras
- VGG16, EfficientNetB0, Decision Tree (DT), Support Vector Machine (SVM), Random Forest (RF), XGBoost (XGB)
- Frontend: Next.js, TailwindCSS, Framer Motion
- Animation & SVG: Custom SVG pipelines and motion-based visualization
- DDTI Dataset: Public ultrasound dataset with XML clinical metadata
- TR12345 Dataset: KMUTT-provided dataset with TIRADS labeled images
DDTI dataset binary classification performance
Classification report and confusion matrix from hybrid model
Best model performance comparison chart
- Hybrid model outperforms pure ML/DL in interpretability and class-level precision
You can view the complete project report, and related artifacts via Google Drive:
git clone https://github.com/Krisha1703/Thyroid-Classification.git
cd thyroid-classification
npm install
npm run devOpen http://localhost:3000 to view in the browser.
Your feedback is highly appreciated to improve this research and simulation.
We also ask for:
- Overall rating of the simulation experience
- Rating of the research methodology presented
We sincerely thank our project advisors and committee members at KMUTT for their invaluable guidance and support throughout the research and simulation development journey.
This project is licensed under the MIT License. See the LICENSE file for details.