Image recognition with missing-features based on gaussian mixture model and graph constrained nonnegative matrix...

Area of Science:

• Medical Image Analysis
• Machine Learning
• Computer Vision

Background:

• The increasing demand for automated medical image recognition highlights the challenge of missing data.
• Missing data is a common and often unavoidable issue in medical imaging applications.

Purpose of the Study:

• To develop a robust method for recognizing medical images containing missing features.
• To address the instability of training Gaussian mixture models (GMMs) directly on high-dimensional data.

Main Methods:

• Feature extraction followed by graph-constrained nonnegative matrix factorization (NMF) for dimensionality reduction.
• Training a GMM on reduced-dimensional data and extending it using alternating expectation conditional maximization (AECM).
• Imputing missing features using marginalizing GMM using Bayesian decision (MGBD) and conditional mean imputation (CMI) during the test phase.

Main Results:

• The proposed GMM-based approach effectively handles missing features in medical images.
• Experimental results on three real datasets validate the feasibility and efficiency of the developed scheme.
• The method demonstrates successful object identification through posterior probability calculation.

Conclusions:

• The developed GMM-based strategy provides a viable solution for medical image recognition with missing data.
• Dimensionality reduction and advanced imputation techniques are crucial for stable and efficient GMM training.
• The proposed method offers a significant advancement in automated medical image analysis.