Deep Belief CNN Feature Representation Based Content Based Image Retrieval for Medical Images

Area of Science:

• Medical Imaging
• Orthopedics
• Artificial Intelligence in Medicine

Background:

• Avascular Necrosis (AN) is a significant cause of musculoskeletal disability, frequently affecting younger populations.
• Early diagnosis and intervention are crucial for managing AN, which commonly impacts the femoral bone, knees, and other joints.
• Retrieving images of AN-affected bones is challenging due to diverse fracture locations and presentations.

Purpose of the Study:

• To propose an effective methodology for retrieving Avascular Necrosis (AN) images.
• To enhance the accuracy and efficiency of AN image retrieval using advanced deep learning techniques.

Main Methods:

• A preprocessing stage involving Median Filter (MF) for noise reduction and image resizing.
• Feature representation using Deep Belief Convolutional Neural Network (DB-CNN) to extract salient image characteristics.
• Transmutation of feature representations into binary codes, followed by similarity measurement using Modified-Hamming Distance for image retrieval.

Main Results:

• The proposed DB-CNN based methodology demonstrated superior performance in retrieving AN images compared to existing techniques.
• Effective noise reduction and feature extraction were achieved through the integrated preprocessing and DB-CNN approach.
• The binary code transformation and Modified-Hamming Distance facilitated accurate similarity computations for retrieval.

Conclusions:

• The developed methodology offers an effective solution for Avascular Necrosis image retrieval.
• This approach holds promise for improving diagnostic capabilities and facilitating timely intervention in AN cases.
• The integration of DB-CNN and Modified-Hamming Distance presents a robust framework for medical image retrieval tasks.