Medical Image Classification for Lower-Extremity Fractures and DDH

NOTE : This file contains describtion for DDH and Lower-Extremity Fracture.

First Model : Lower-Extremity Fracture .

What does it contain?

Environment Preparation Installation of libraries. Import of tools. Selection of GPU.

Data Preparation

Includes:

Image Transformations CLAHE Data Augmentation 3. Classification Model Building

Based on:

DenseNet121

For classifying radiographic images.

Model Training

Includes:

Training Validation Loss Curves 5. Model Evaluation

Includes:

Classification Report Confusion Matrix 6. Interpreting Results Using Grad-CAM

Displays the regions the model relied on during decision-making.

This helps in interpreting the AI ​​results.

Using YOLO

The YOLO model was used to detect injury areas within the radiographic image.

Injury Severity Assessment

This section includes logic for determining injury severity based on model results.

Final Report

This section includes:

Classification results.

YOLO results.

Injury severity.

Medical advice.

Purpose of the document

This document represents the initial version in which most of the project's concepts, such as:

Classification GradCAM YOLO Detection Severity Assessment, were tested and developed before being integrated and refined into the final version.

Second Model : DDH .

Brief Description

This file contains a comprehensive system for diagnosing Developmental Dysplasia of the Hip (DDH) using artificial intelligence and deep learning techniques. It begins with classifying radiographic images, then identifies keypoints, calculates the severity of the injury, and finally provides a report and medical guidance.

What's Included?

Environment Setup Installation of required libraries. Import of all libraries. Automatic selection of GPU or CPU. Linking to Google Drive when using Google Colab.

Image Preprocessing

Includes:

CLAHE for contrast enhancement of radiographic images.

Image Normalization.

Image Transformations.

Gaussian Heatmaps for Keypoint Training. 3. Classification Model

Based on:

DenseNet121

Used to classify X-rays into:

Normal DDH

The model is trained and evaluated using:

Accuracy Precision Recall F1-score Test-Time Augmentation (TTA)

Keypoint Detection Model

Includes a model based on:

DenseNet121 Encoder Heatmap Decoder

Identifies keypoints specific to the hip joint.

It also includes:

Soft-Argmax Heatmap Loss PCK Metric OKS Metric 5. Medical Measurements

Calculates important indicators such as:

Acetabular Index (AI) Anatomical measurements necessary to determine the severity of injury.

Severity Assessment

This relies on measurements extracted from keypoints to determine the severity of the injury.

Medical Guidance

After determining the severity of the injury, the system displays:

Severity level Appropriate medical recommendations A simplified interpretation of the result 8. Model Training

The file contains:

Data Loading Data Segmentation DataLoaders Training Loops Validation Evaluation Plotting training curves 9. Full Diagnosis Pipeline

Finally, there is a complete pipeline that:

Loads the X-ray image

Classifies it Extracts keypoints Calculates measurements Determines the severity of the injury Displays the final diagnostic report

Purpose of the file

This file represents the complete, integrated system for the project, which combines all stages of diagnosis within a single notebook.