Segmented MR Images by RG-FCM subjected to Non-Uniform Compression comprising Cascade of different Encoders

Area of Science:

• Medical Imaging
• Image Compression
• Machine Learning

Background:

• Medical image transmission and storage face challenges due to large file sizes and data redundancy.
• Reducing medical image size without losing diagnostic information is crucial with the growth of digital imaging data.
• Existing compression methods often rely on manual or traditional segmentation, limiting efficiency.

Purpose of the Study:

• To enhance medical image compression efficiency through accurate segmentation and non-uniform compression.
• To develop an unsupervised machine learning approach for precise extraction of informative image regions.
• To improve the performance of medical image compression techniques.

Main Methods:

• Proposed an unsupervised modified fuzzy c-means (FCM) clustering for segmentation.
• Integrated an automated region-growing algorithm with FCM (RG-FCM) to handle noise and improve segmentation.
• Applied a cascade of encoders with differential bit rates for informative and background image regions.

Main Results:

• The proposed RG-FCM technique demonstrated superior segmentation performance on Magnetic Resonance Imaging (MRI) datasets.
• The integrated segmentation and non-uniform compression approach achieved higher compression metrics compared to existing methods.
• Empirical analysis confirmed the technique's effectiveness in improving both segmentation and compression.

Conclusions:

• The combination of segmentation techniques improved Jaccard and Dice indexes, validating the proposed compression method.
• The cascade of encoders further endorsed the superior performance of the novel compression technique.
• This method enables higher compression of medical images while preserving clinically significant information.