iraqigold/lung-cancer-segmentation-vision

AI-Based Intelligent Extraction of Lung Cancer Tumors (Segmentation)

1. Research Overview

This repository contains a preprocessed, standardized, and ready-to-use version of medical radiography images for Lung Cancer Segmentation. This dataset serves as a core computational component for our research project: "AI-Based Intelligent Extraction of Diseased Areas from Medical Radiography Images".

2. Source & Attribution

The raw data was originally sourced from the Kaggle platform. We acknowledge and thank the original authors for providing the foundational CT slices and mask annotations. 🔗 Original Kaggle Dataset: Lung Tumor Segmentation

3. Dataset Methodology & Refinement

The original dataset was highly structured but fragmented across numerous subject-specific subfolders, which poses a challenge for direct continuous training in segmentation networks. To optimize this data for deep learning architectures (specifically U-Net), we applied a custom Python preprocessing pipeline:

• Flattened Hierarchical Folders: We merged the scattered subject directories into a unified, flat images/ and masks/ structure for seamless batch loading.
• Unique Traceable Identification: We implemented a robust algorithmic naming convention ({split}_{subject}_{filename}.png). This ensures full traceability back to the original patient slice and completely eliminates file-overwriting conflicts.
• Ready-to-Train Format: All inputs and ground-truth labels are standardized in lossless PNG format, ensuring exact 1:1 pixel mapping between the CT scan and the pathological mask.

4. Repository Structure

• /images: Contains all 2D CT slices of the lungs.
• /masks: Contains the corresponding binary ground-truth labels (Tumor regions are explicitly segmented).
• lung-cancer-vision-v1.zip: A compressed, all-in-one archive of the entire processed dataset for rapid download and cloud deployment.

5. Clinical Application

Unlike basic classification datasets that only predict the presence of a disease, this pixel-level segmentation data trains AI models to understand the exact spatial morphology, size, and boundaries of lung nodules, directly assisting in rapid and automated clinical diagnostics.