Fat-based studies for computer-assisted screening of child obesity using thermal imaging based on deep learning techniques: a comparison with quantum machine learning approach

• Published in: Soft computing (Berlin, Germany) Vol. 27; no. 18; pp. 13093 - 13114
• Main Authors: Rashmi, Richa, Snekhalatha, U., Krishnan, Palani Thanaraj, Dhanraj, Varun
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01-09-2023; Springer Nature B.V
• Subjects: Abdomen; Accuracy; Adipose tissue; Artificial Intelligence; Artificial neural networks; Body fat; Body mass index; Children; Computational Intelligence; Computing costs; Control; Customization; Deep learning; Engineering; Focus; Heat detection; High temperature; Image classification; Machine learning; Mathematical Logic and Foundations; Mechatronics; Medical imaging; Obesity; Quantum computing; Regional analysis; Regional development; Robotics; Screening; Temperature gradients; Thermal imaging; Deep learning; Obesity; Variation quantum classifier; Childhood obesity; Convolution neural network; Thermal imaging; Quantum machine learning
• ISSN: 1432-7643; 1433-7479
• DOI: 10.1007/s00500-021-06668-3

The main objectives are (i) to study the relation of temperature of brown adipose tissue (BAT) with respect to obesity in different regions of the human body and to predict the most significant regional thermogram in determining the obese condition in children; (ii) to develop an customized convolutional neural network (CNN) model and to compare its performance with the modified pre-trained models for the automated screening of childhood obesity; (iii) to explore the application of quantum computing, namely variational quantum classifier (VQC), in the image classification of obesity condition. The mean skin surface temperature was measured in neck, forearm, abdomen and calf region for normal, overweight and obese children. Two modified pre-trained CNN (MobileNet-V2 and VGG-16) and Customized Net are studied for multi-class categorization of normal, overweight and obese thermograms. In addition, VQC was simulated and compared with the performance of deep learning techniques. Among the regional temperature studied, abdominal region exhibited high temperature difference of 3.72 °C (10.96%) and 5.56 °C (16.84%) between normal vs overweight and normal vs obese, respectively. The proposed Customized Net model achieved the best overall accuracy of 91.6% and 89.3% for abdomen region and neck region, respectively. Similarly, VQC has illustrated best classification accuracy in neck (84.4%) and abdomen (82.2%) regions. Moreover, VQC model promotes reduction in the computational cost for the diagnosis of obesity based on thermal imaging. Hence, regional thermogram assessment with deep learning and quantum computing approach is considered as a feasible method for preliminary screening of overweight and obesity condition in children.