Chronic Kidney Disease (CKD) Multimodal Dataset

📌 Project Overview

This project builds a multimodal dataset for Chronic Kidney Disease detection. It combines:

1. Tabular Clinical Data: Blood tests (Creatinine, Hemoglobin, etc.) and patient metadata.
2. Medical Imaging: CT Scans of kidneys (Normal, Cyst, Tumor, Stone).

The goal is to create a dataset where every patient record (CSV row) is paired with a corresponding medical image, allowing for multi-input machine learning models (e.g., combining MLP for data and CNN for images).

📂 File Structure & Functionality

1. img_download.py (Step 1)

• Purpose: Downloads the CT Kidney dataset from Kaggle.
• Action: Uses the Kaggle API to fetch the dataset, extracts it, and reorganizes the folders into a clean structure (kidney_images/Tumor, kidney_images/Normal, etc.).

2. ds_generate.py (Step 2)

• Purpose: Generates the clinical CSV file.
• Action:
  • Downloads real clinical data from a remote repository.
  • Cleans the data (handling missing values, fixing types).
  • Augments the dataset to 1500 rows using statistical sampling and noise injection to ensure enough data for training.
  • Assigns a theoretical image path to each row based on medical logic (e.g., High Creatinine -> Tumor image).

3. fix_path.py (Step 3)

• Purpose: Connects the CSV to the actual downloaded images.
• Action:
  • Scans the kidney_images folder to find all valid .jpg files.
  • Updates the CSV generated in Step 2.
  • Replaces the theoretical paths with actual file paths that exist on your disk, ensuring no "File Not Found" errors during training.

4. audit_data.py (Step 4)

• Purpose: Quality Assurance.
• Action:
  • Checks class balance (Sick vs. Healthy).
  • Verifies that every image path in the CSV actually exists.
  • Runs a baseline Logistic Regression model on the tabular data to estimate the "information power" of the blood test features.

🚀 How to Run

Execute the scripts in this specific order to build the dataset:

1. Download Images:

   python img_download.py

2. Generate Clinical Data:

   python ds_generate.py

3. Link Images to Data:

   python fix_path.py

4. Verify Integrity:

   python audit_data.py

---

📊 Data Audit Report

(Generated from audit_data.py)

🕵️ STARTING DATA AUDIT...

✅ CSV Loaded: 1500 rows

📊 CLASS DISTRIBUTION:

• CKD (Sick): 405 (27.0%)
• Normal (Healthy): 1095 (73.0%) Note: A 'dumb' model guessing the majority class would have 73.0% accuracy.

🔍 DATA QUALITY:

• Missing Values: 0
• Duplicate Rows: 0

🖼️ IMAGE INTEGRITY SCAN:

• Images Found: 1500
• Images Missing: 0
• Corrupt Files: 0

======================================== 📈 ESTIMATED CLINICAL METRICS (Based on Blood Work) (Running rapid Logistic Regression on Tabular Data...)

✅ Accuracy: 0.9967 🎯 Precision: 1.0000 🔍 Recall: 0.9882 🛡️ Specificity: 1.0000 📉 RMSE: 0.0557

Interpretation: These numbers represent the 'Information Power' of your dataset. If a simple statistical test gets high accuracy, your data is very clean.