Tumor Malignancy Detection Using Histopathology Imaging.

Area of Science:

• Biomedical Imaging
• Computational Pathology
• Digital Health

Background:

• Accurate cancer diagnosis and grading rely heavily on image segmentation and classification in biomedical imaging.
• Histopathology images provide crucial information for cancer assessment, necessitating advanced analytical techniques.

Purpose of the Study:

• To develop and evaluate a novel method for classifying histopathology images for improved cancer diagnosis and grading.
• To combine handcrafted and shape features using bag of visual words (BoW) for robust image classification.

Main Methods:

• A multistage segmentation technique was employed to localize nuclei in histopathology images, involving stain decomposition and histogram equalization.
• Nuclei key points were extracted using the fast radial symmetry transform, followed by normalized graph cut for region estimation and modified gradient for boundary estimation.
• Handcrafted features from localized nuclei regions and shape features using bag of visual words (BoW) were extracted for classification.

Main Results:

• The proposed method achieved high classification accuracies on the Bisque and BreakHis datasets, reaching 93.87% and 96.96%, respectively.
• Experiments demonstrated the advantage of combining local nuclei features (handcrafted) and global spatial features (BoW).

Conclusions:

• The proposed method, integrating handcrafted and BoW features, significantly enhances the performance of cancer diagnosis and grading using histopathology images.
• This approach offers a promising tool for improving diagnostic accuracy in computational pathology.