KBRT-HNSCC

Knowledge-Based Radiotherapy for Head and Neck Squamous Cell Carcinoma (HNSCC)

Abstract

Head and neck squamous cell carcinoma (HNSCC), accounting for 90% of head and neck cancers (HNC) and ranking as the 6th most common globally, exhibits an inherent resistance to conventional radiotherapy. The development of advanced strategies, particularly leveraging Machine Learning (ML), is crucial for optimizing HNSCC treatment. This paper presents the foundational work for an ML-driven approach to predict optimal radiotherapy dosages for HNSCC patients. Our initial investigation focused on evaluating and identifying the most effective ML models for classifying HNSCC patient data, a critical precursor to accurate dosage prediction. We systematically assessed several supervised learning algorithms, to include, binary and multi-class Log Regression, SVM, CNN, ResNet, and U-Net, using a comprehensive dataset. The results of this comparative analysis pinpoint different forms of CNN as the most promising for HNSCC data classification, demonstrating higher accuracy throughout. This foundational work establishes the classification framework necessary for subsequent stages of personalized radiotherapy optimization.

Overview

This project explores the application of knowledge-based radiotherapy (KBRT) models to improve treatment planning for head and neck cancer (specifically HNSCC). By combining mathematical modeling and machine learning, we aim to optimize dose distribution and reduce treatment-related toxicity, ultimately improving patient outcomes.

🧑‍🚀 Team

We are an interdisciplinary student research team bringing together expertise in mathematics, bioinformatics, computer science, and data science.

Somiya Rauf

PhD candidate in Mathematics w/ Concentration in Bioinformatics.
Experienced in mathematical modeling of cancer immunotherapy, radiotherapy, and chemotherapy using MATLAB and Python. Currently expanding into machine learning to enhance research methods and integrate data-driven insights. Github

---

Abdul-Malik Mohammed

Computer Science - Senior, final semester
Skilled in training ML models and SWE applications during internships and industry projects. Proficient in Python (PyTorch), TypeScript, React, Express, and Node.js. Github

---

Joshua Pina

Data Science - Senior, plans to pursue JD + MS in Analytics.
Background in Python, applied probability & statistics, data analytics, big data programming, and ML. Experienced Program Manager in government contracting and former U.S. Army Combat Medic. Github

---

Goals

• Develop and validate a KBRT model specifically for HNSCC cases.
• Compare model-predicted vs. actual clinical dose distributions.
• Explore integrating ML-based predictions with traditional mechanistic models.
• Share reproducible code and visualizations for the research community.

Full Tech Stack

• Python
  • Scikit-learn
  • PyTorch
  • NumPy
  • Pandas
  • Matplotlib
  • Seaborn
  • XGBoost
• MATLAB
• Streamlit

Consideration for using Azure, and for future expansion, converting Streamlit to a full-stack application.

Development Environment

This project uses GitHub Codespaces to ensure a consistent, reproducible development setup.

---

Quick Start

1. Open in Codespaces
   Click the green Code button on this repository, then choose Open with Codespaces → New codespace.

2. Automatic setup
   The dev container will automatically install Python and all required packages listed in requirements.txt.

3. Activate the environment
   When your Codespace starts, you’re ready to run scripts and notebooks immediately.

---

Project Website

👉 View website

---

Updated view