Estimation of tumor parameters using neural networks for inverse bioheat problem
Introduction
Cancer is a broad term describing the disease that is a consequence of cellular changes causing uncontrolled growth and division of cells. Certain types of cancer cause rapid cell growth, while others cause cells to grow and divide at a slower rate. Certain forms of cancer result in visible growths called tumors, while others, such as leukemia [1]. Breast cancer starts in the breast, usually in the inner lining of the milk ducts or lobules of the breast. There are various types of breast cancer, having different stages, aggressiveness, and genetic makeup. Treatment includes surgery, radiation, and drugs (hormone therapy and chemotherapy) [2].
Cancer diagnosis, at its earliest stages, often provides better chances for a cure. During a physical exam for diagnosis, the doctor may look for abnormalities like changes in the skin colour or enlargement of an organ that may perhaps indicate the presence of cancer. Laboratory tests, such as urine and blood tests, may also help doctors identify abnormalities that can be caused by cancer. Imaging tests are a non-invasive way to diagnose cancer and include bone scintigraphy, computerized tomography (CT) scan, positron emission tomography (PET) scan, magnetic resonance imaging (MRI), ultrasound, and X-ray, among others. During a biopsy, the doctor collects a sample of cells for testing the presence of cancerous cells in the laboratory [3].
A significant limitation of the current approaches is that the subjective morphological examination of cytological specimens alone is highly prone to error. It becomes notoriously difficult to differentiate between normal and dysplastic cells at an early stage of malignant transformation. Also, the use of intrusive methods to estimate the size and position of the tumor is a relatively cumbersome and annoying process. Non-intrusive methods have an advantage in this aspect [4].
A neural network is a set of algorithms that seeks to understand underlying relationships in a dataset through methods that closely resemble how the human brain works. Neural networks model itself based on the various intricacies present in the dataset so the results that the network generates do not have to change according to the changes in data rather it takes into account the changes and tries to create the best possible output based on the parameters that are used while designing the network. A “neuron” in a neural network is a mathematical function that processes and tries to find correlations between the various aspects of the data that enters it. The Neural network finds a correlation between the input and output signals. Hidden layers refine and tweak the input weights until the neural network's margin of error reaches close to acceptable levels throughout training the model. It is hypothesized that hidden layers give more importance to certain features in the input data that have a greater say on the output that is predicted [5]. For the experiment that is undertaken in this paper, there is a need to find patterns from the thermograph readings of the tumor. Since it is a task that involves the usage of vast amounts of data and finding patterns, applying neural networks would be the best option to get the desired results.
Pennes [6] proposed a simplified bioheat model describing the effects of metabolic heat generation and blood perfusion on heat transfer within living tissues. Kumar et al. [7] use the Pennes equation to govern bioheat transfer within the mathematical model of a breast tissue embedded with a tumor. Ströher et al. [8] advocate the use of heat as a powerful tool in diagnosing cancers, with an added advantage of being a non-intrusive or non-invasive method of examination. N. M. Sudharsan et al. [9] found that the surface temperature distribution of the breast provides information on the presence of a tumor. This distribution has a relation to the size and location of the tumor and can be seen using thermography, where the infrared radiation emitted from the surface of the breast is recorded and a thermal pattern obtained. Subhadeep et al. [10] proposed the estimation of the position and size of a spherical tumor in a human breast using temperatures obtained on the surface of the breast through a breast thermogram in conjunction with artificial neural networks. Agnelli et al. [11] have presented an estimation methodology to determine the unknown thermophysical or geometrical parameters of the tumor region using the temperature profile on the skin surface obtained by infrared thermography. The bioheat Pennes equation is solved for and the parameters are determined from the temperature profiles obtained from simulated data. Umadevi et al. [12] proposed a framework for estimation of tumor size using clever algorithms and the radiative heat transfer model. They incorporated the more realistic Pennes bio-heat transfer model and Markov Chain Monte Carlo (MCMC) method and established the capacity of this method to give quick, dependable results against erroneous data in tune with the accuracy standards today. Mital et al. [13] presented a framework to determine the breast tumor parameters using the profile of surface temperature obtained by infrared thermography. The estimation methodology involves evolutionary algorithms with an artificial neural network (ANN) and genetic algorithm (GA) model. Convergence plots of the mean squared error of the neural network output during training were obtained. Kumar et al. [7] estimated the unknown heat flux at the surface of a solid body. Hafid et al. [14] aimed to develop an inverse heat transfer approach to predict the time-varying freezing front and temperature distribution of tumors during cryosurgery. The direct model rests on a one-dimensional Pennes bioheat equation. Recently some interesting works were published in field of application of Neural Networks which can be interesting for future works [15], [16], [17], [18].
This work proposes to solve the problem of intrusive canker tumor detection by using a method that can decide the cancer tumor parameters from the temperature profile of the cancerous tissue. The process involves the generation of cancer tumor models that are based on the penne's bioheat equation and using this data to train a neural network model that can detect the cancer tumor parameters from the temperature profile distribution from the surface of the body.
Section snippets
Problem statement
A 3D model of the human breast embedded with a spherical tumor is simulated according to the Pennes bioheat transfer equation. The temperature distribution of the breast tissue from the simulation generated serves as the basis of data for training the ANN model. Multiple training cases are simulated by varying the size and position of the tumor inside the breast to generate sufficient data for training the neural network created in the present work. The optimum neuron count in the hidden layer
Modelling of breast tissue and material selection
The software chosen for modeling and simulation purposes is COMSOL Multiphysics® and the parameters such as shape, size, and material of the base model are decided based on the following reasons. Anderson et al. [19] have reported that the average woman in almost every country, except in the US, has pear-shaped (semi-ellipsoid) breasts rather than the conventional round (hemispherical) shape. Contijoch et al. [20] also advocate ellipsoid fitting in the shape analysis of the anatomy of the
RMSE values and the selection of hidden layer neuron
The simulations using the neural networks had found out that the Root Mean Square Error is very high when the neurons are less and it decreases with increasing the number of neurons in the hidden layer. There is a general trend of decreasing Root Square Mean Error with the increase in the number of neurons until the optimum number of neurons with the least amount of error is reached, except for the odd outlier in the hidden layer. In the simulation, the findings show that the least error is
Conclusion
The trained model was tested on 7 entirely different models and simulations. The results have been tabulated in Table 2 and the following observations are made. For models embedded with tumors whose size lies in categories T1a, the prediction of tumor parameters is noticeably different from the actual parameters. This is especially the case when such tumors are located in breasts of comparatively larger sizes. However, for tumors larger than 0.5cm (size category > T1a), the model predicts the
Declaration of Competing Interest
There is not any conflict of interest for all authors of this manuscript.
Acknowledgments
The authors would like to thank Deanship of Scientific Research at Majmaah University for supporting this work under Project Number No. R- 2021-79.
References (24)
- et al.
A neural network-based estimation of tumor parameters from a breast thermogram
Int. J. Heat Mass Transf.
(2010) - et al.
Tumor location and parameter estimation by thermography
Math. Comput. Modell.
(2011) - et al.
Fast inverse prediction of the freezing front in cryosurgery
J. Therm. Biol.
(2017) - et al.
ANN: Prediction of an experimental heat transfer analysis of concentric tube heat exchanger with corrugated inner tubes
Appl. Therm. Eng.
(2017) - et al.
Effects of the preparation technique on the physicochemical characteristics and dissolution improvement of ketoprofen-SBE7-β-CD binary inclusion complexes
Colloids Surf. A
(2021) - et al.
A new predictive equation for resting energy expenditure in healthy individuals
Am. J. Clin. Nutr.
(1990) - Cancer: Overview, causes, treatments, and types. (n.d.). Retrieved November 15, 2020, from...
Novel anticancer agents for treatment of breast cancer utilizing in silico optimization and comparability to Melphalan
Cancer Res. J.
(2011)Cancer
(2018, December 12)- et al.
New minimally invasive approaches to early cancer detection
Eur. Oncol. Haematol.
(2007)
An innovative neural network approach for stock market prediction
J. Supercomput.
Analysis of tissue and arterial blood temperatures in the resting human forearm
J. Appl. Physiol.
Cited by (6)
Deep learning-based approach in surface thermography for inverse estimation of breast tumor size
2024, Scientific AfricanModeling of glioma growth using modified reaction-diffusion equation on brain MR images
2022, Computer Methods and Programs in BiomedicineCitation Excerpt :But, in existing studies of brain tumor growth based on reaction-diffusion models, the diffusion coefficients of tumors are simply differentiated into WM, GM and CSF, while the proliferation coefficients of tumors are globally parameterized and fail to differentiate in detail according to the different regions of the tumor. It is difficult for these method to characterize spatial variation of the diffusion and proliferation coefficients [21–23,33]. We calculated diffusion and proliferation coefficients for prediction of glioma growth based on patient-specific anatomy.
CNN-Based Approach for Non-Invasive Estimation of Breast Tumor Size and Location Using Thermographic Images
2024, International journal of online and biomedical engineering3D-BreastNet: A Self-supervised Deep Learning Network for Reconstruction of 3D Breast Surface from 2D Thermal Images
2023, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)