Extracting Regions of Interest and Selective Feature Application in Leukaemia Image Classification.

Area of Science:

• Medical imaging analysis
• Computational pathology
• Machine learning in diagnostics

Background:

• Accurate leukaemia diagnosis from blood smear images is crucial for effective treatment.
• Morphological changes in white blood cells offer key features for leukaemia detection.
• Limited datasets can hinder the performance of machine learning models like CNNs.

Purpose of the Study:

• To explore methods for improving leukaemia classification accuracy using reduced datasets.
• To investigate the impact of feature extraction and Convolutional Neural Network (CNN) application on image classification.
• To evaluate the effectiveness of extracting Regions of Interest (ROI) for enhancing CNN performance.

Main Methods:

• Utilized Otsu thresholding to segment initial leukaemia images into Regions of Interest (ROI).
• Created two datasets: original images and segmented ROI images.
• Applied feature extraction techniques and Convolutional Neural Networks (CNNs) for classification analysis on both datasets.

Main Results:

• Feature extraction demonstrated superior performance on the original dataset.
• CNN classification accuracy was significantly higher when applied to the ROI dataset.
• The study identified ROI extraction as a key factor for improving CNN performance in leukaemia detection.

Conclusions:

• Extracting Regions of Interest (ROI) from leukaemia images enhances CNN classification accuracy.
• Further refinement by filtering specific ROI can lead to more precise feature extraction and improved diagnostic accuracy.
• This approach offers a promising direction for developing more robust automated leukaemia detection systems.