Fuzzy modelling
M. Sweetline Sonia; S. Sumathi
Volume 23, Issue 6 , November and December 2026, , Pages 139-156
Abstract
Early detection of cerebral palsy (CP) is challenging due to subtle and heterogeneous neuromotor abnormalities in infants. This paper proposes FLinkNet-GAN, a hybrid deep learning and fuzzy logic framework for membership-based CP risk assessment from RGB-D infant videos. Initially, a Bright ...
Read More
Early detection of cerebral palsy (CP) is challenging due to subtle and heterogeneous neuromotor abnormalities in infants. This paper proposes FLinkNet-GAN, a hybrid deep learning and fuzzy logic framework for membership-based CP risk assessment from RGB-D infant videos. Initially, a Bright Contrast Dynamic Histogram Equalization (BCDHE) filter enhances motion visibility. A Generative Adversarial Network (GAN) performs multi-pose estimation to extract kinematic features, including Movement Deviation (MD) and Joint Trajectory Discontinuity (JD). These features are processed through a LinkNet backbone to preserve fine spatial details and subsequently mapped into a Mamdani-type Fuzzy Inference System using Gaussian membership functions. Unlike conventional hard classifiers, the proposed approach generates a continuous Risk Score (RS) via centroid defuzzification, providing interpretable risk levels: Normal, Borderline, High, and Very High. Experimental evaluation on the RVI-38 dataset achieves 94.85% accuracy and 94.5% pose precision, demonstrating robust and uncertainty-aware CP risk stratification.