A DEEP SIMILARITY LEARNING FRAMEWORK FOR AUTOMATED VISUOSPATIAL COGNITIVE ASSESSMENT AND EARLY COGNITIVE IMPAIRMENT SCREENING USING SIAMESE NETWORK
DOI:
https://doi.org/10.52152/vwqpj653Ključne besede:
Visuospatial Assessment, cognitive, similarity calculation, deep learning, ResNet50Povzetek
The brain network which is responsible for visuospatial abilities is mainly controlled by the parietal and occipital lobes. The damages in the occipital lobe represents the individual’s vision disability which fails to recognize the objects. The damages in the parietal lobe causes an inability to interact with or manipulate objects in the daily activities. Existing visuospatial conventional approaches suffer from clinician subjectivity, lower reproducibility, time consuming, manual assessment and scoring. This guarantees the development of the automated assessment tools for visuospatial ability assessment. Our proposed Siamese backbone ResNet50 approach evaluates the individuals drawing like infinity loop and 3D cube by comparing with the reference images, then automatically award the marks based on the image similarity scores. This score helps to identify the cognitive impairment at the early stage. Experimental results demonstrate that the proposed approach provides accurate, consistent and fine-grained visuospatial assessment, outperforming traditional subjective scoring methods. The proposed Siamese with ResNet50 network provides a scalable and reliable solution for early cognitive screening, longitudinal observing and remote assessment of visuospatial deficits.
Prenosi
Objavljeno
Številka
Rubrika
Licenca
Avtorske pravice (c) 2026 Lex localis - Journal of Local Self-Government

To delo je licencirano pod Creative Commons Priznanje avtorstva-Nekomercialno-Brez predelav 4.0 mednarodno licenco.


