🦴 Pediatric Bone Age Detection AI 🏥

📌 Overview

This is an open-source AI model for predicting bone age from pediatric X-ray images, based on the RSNA Bone Age dataset. The model uses DeepLabV3+ for segmentation and EfficientNetV2-M for bone age regression. It is optimized with techniques like test-time augmentation (TTA).

The model is deployed via a FastAPI backend and is accessible through an API, which is integrated with a React web app.

---

⚙️ Features

✅ Bone age prediction from X-ray images
✅ Automated segmentation with DeepLabV3+
✅ EfficientNet-V2M for analysis
✅ Test-time augmentation for better accuracy
✅ Supports inference via API (FastAPI-based)
✅ Optimized for Google Colab (uses A100 GPU)
✅ Fully open-source under OpenRAIL License

---

🏗 Model Architecture

1️⃣ Image Segmentation

• Model: DeepLabV3+ (ResNet-50 backbone)
• Purpose: Extracts the region of interest (bones)

2️⃣ Bone Age Regression

• Model: EfficientNetV2-M
• Additional Features: Gender input as auxiliary feature
• Training:
  • 10,000 training images (RSNA dataset)
  • 1,200 validation images
  • Augmentations

3️⃣ Calibration Model

• Linear regression layer for final calibration

---

🚀 API Usage

This model is deployed via a FastAPI backend on Hugging Face Spaces.

Health Check

curl -X GET "https://ameyakawthalkar-boneagealpha.hf.space/health"