Effective CBMIR System Using Hybrid Features-Based Independent Condensed Nearest Neighbor Model.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Informatics

Background:

• Increasing volume of medical images necessitates efficient retrieval systems.
• Variations in image shape and size pose challenges for large medical databases.
• Automated systems are crucial for effective medical image retrieval.

Purpose of the Study:

• To develop an automated system for enhanced medical image retrieval.
• To improve the accuracy and efficiency of retrieving Pap smear cell images.
• To reduce the semantic gap in medical image feature representation.

Main Methods:

• Acquired medical images from a new Pap smear dataset.
• Applied image normalization to enhance visible quality.
• Utilized hybrid feature extraction: Histogram of Oriented Gradients (HOG) and Modified Local Binary Pattern (MLBP).
• Employed an Independent Condensed Nearest Neighbor (ICNN) classifier for seven cell image classes.
• Implemented chi-square distance measure for relevant image retrieval.

Main Results:

• The proposed hybrid feature extraction effectively reduced the semantic gap.
• The ICNN classifier accurately classified seven classes of cell images.
• Achieved high performance metrics: specificity, recall, precision, accuracy, and F-score.
• Attained a retrieval accuracy of 98.88%, surpassing deep learning models (LSTM, DNN, CNN).

Conclusions:

• The developed automated system demonstrates superior performance in medical image retrieval.
• The hybrid feature extraction approach is effective for Pap smear image analysis.
• The proposed model offers a significant improvement over existing deep learning methods for this task.