Discriminative generalized Hough transform for object localization in medical images

Area of Science:

• Medical image analysis
• Computer vision
• Machine learning

Background:

• Object localization is crucial for medical image analysis.
• Traditional methods may lack efficiency and robustness.
• The generalized Hough transform (GHT) is a foundational technique.

Purpose of the Study:

• Propose the discriminative generalized Hough transform (DGHT) for efficient and reliable medical image object localization.
• Provide theoretical insights, methodological overview, and application scope.
• Enhance robustness and accuracy in object detection.

Main Methods:

• Combines generalized Hough transform (GHT) with discriminative training.
• Assigns individual weights to model points, trained to minimize localization error.
• Incorporates extensions for automatic model generation, iterative refinement, joint model training, and multi-level processing.

Main Results:

• Achieved 97.6% success rate for knee localization in long-leg radiographs.
• Demonstrated 95.5% success rate for vertebrae localization in C-arm CT.
• Reached 100% success rate for femoral head localization in whole-body MR.
• Outperformed Hough forests for knee localization with a 97.8% success rate.

Conclusions:

• DGHT is a versatile and effective procedure for medical image object localization.
• The method demonstrates high success rates across diverse imaging modalities and anatomical targets.
• DGHT offers a robust and efficient solution for automated medical image analysis tasks.