Authors:
P. Vidhya, S. Silvia Priscila

Addresses:
Department of Computer Science, Bharath Institute of Higher Education and Research, Chennai, Tamil Nadu, India. 

Abstract:

The early diagnosis of Down syndrome is relevant in the aspect of medical care and developmental provisions. The study proposes a novel multimodal deep learning model that combines craniofacial anatomy and thermal image patterns to identify the condition in children. The model is a mixture of structural dysmorphism and physiological heat signatures, enabling it to sensitize to small signals that unimodal systems would otherwise ignore. The experiment is grounded in an edited set of 458 data items, which include high-resolution 2D images of faces and infrared thermograms of faces. It was developed in Python using the TensorFlow/Keras framework, with pre-trained feature-extraction architectures and custom fusion layers for combining and classifying features. Experimental findings indicate that sensitivity and specificity are high, suggesting that the non-invasive test combining thermal and visual data is highly effective in improving diagnostic accuracy. This framework will provide a reproducible, low-cost screening tool that will help minimize the use of invasive genetic testing in the initial assessment.

Keywords: Down Syndrome; Multimodal Fusion; Thermal Imaging; Craniofacial Analysis; Deep Learning (DL); Invasive Genetic Testing; Classifying Features; Diagnostic Accuracy.

Received on: 24/05/2025, Revised on: 13/08/2025, Accepted on: 06/09/2025, Published on: 29/06/2026

DOI: 10.69888/FTSCL.2026.000702

FMDB Transactions on Sustainable Computer Letters, 2026 Vol. 4 No. 2, Pages: 119-128

• Views : 49
• Downloads : 11

Download PDF