Development of methodology for detection of abnormal functioning region in human external organs using thermogram analysis

Abstract

This study has established thermography as a potential diagnostic tool for detecting and parameterizing tumors even at the earlier stage from abnormal local surface thermal features captured by high sensitive infrared cameras without known risk of morbidity. Discrepant thermal features originate not only for tumor’s distinguishing physio-bio-thermal features but also for the state of health, resulting in thermography as a useful tool for retrieving information about the tissue’s health, thus an efficient screening tool. Accurate linking between hyper functional tissues and thermal pattern could turn the screening tool into a promising parameterizing tool. Human external organs, for example chest, forearm and breast have been modeled, mimicking their shapes, inhomogeneity and deformations to obtain steady-state thermal feature in the tissue interior at healthy state and the computation is extended for tumors buried into healthy tissues for determining abnormal local surface thermal image. Tumor diagnosis parameters have been forecasted from thermogram using an indirect process involving the optimization process. The study has applied gradient (gradient projection method), non-gradient (pattern search method) and learning based (genetic algorithm) optimization approaches. Feasibility of the proposed technique is investigated for tumors in human organs. The local abnormal thermal feature screens earlier stage tumors out and reveal how tumors affect the thermal behaviour and what particular parameters have significant influence on the thermal image. Influential parameters are applied as optimization variables and their influences are also figured out to determine the gradient matrix for the gradient optimization technique. The study has employed bio-heat equations, heat-source model and Artificial Neural Network as governing equation to develop simulated datasets. The simulated dataset is compared with test thermogram to minimize a cost function. In lieu of clinical thermograms, the study has developed pretend thermogram with enveloping the simulated datasets with ±10% random noise. This research has tailored optimization algorithms for estimating tumor depth, size, blood perfusion rate, thermal conductivity, and metabolism and the obtained results show good accuracy. The estimated parameters are given to a trained network to reconstruct the thermal feature, thus, validates the performance of the proposed methodology.