Executive Summary : | In the project, machine learning algorithm will be applied for detection and monitoring of early growth of the malignant tissues inside the human breast model based on the scattering parameters. Two ultra-wide band (UWB) antennas will be designed and placed on the air-breast phantom interface for acquiring the transmission and reflection coefficients for detection of healthy, non-malignant and malignant tissues at different frequency bands ranging from 1-12 GHz. A breast phantom model consisting of skin, fat, gland and tumor will be prepared. The dielectric properties of the experimental breast phantom model will be characterized for the different sizes and locations of the benign and malignant tumours inside the breast model. Based on the dielectric medium inside the breast model statistical variation of scattering parameters will be classified using ML algorithms for classification of healthy and malignant breast tissues. The study will include the comparative analysis of the performances of the algorithms: Support Vector Machine (SVM,) Naive Bayes, Decision Tree, K-Nearest Neighbours (KNN) and Neural Network (NN) algorithms for detection of malignant tissues. Due to the transmission of the electromagnetic waves through the breast region, the transmission (S21) and the reflection (S11) coefficients will be varied and the features of S21 and S11 for the wide range of frequencies will be recorded. The project will classify the benign and malignant cells present in the breast model using temporal, spectral, and statistical domain features obtained from the scattering signals. The algorithms will exploit suitable mapping function on each data instance to map the original nonlinear observations into a higher-dimensional space in which they will become separable. For classification the characterization capability of nonlinear features like high order statistics and cumulants and nonlinear feature reduction methods will be combined with linear features. For feature extraction, the energy distribution, frequency content, entropy, energy ratio variance (ERV), energy ratio skew (ERS) and its cumulative values, instantaneous mean frequency (IF) and squared instantaneous bandwidth (IBS) and instantaneous energy (IE) will be analysed. The project will study the dynamic changes taking places in the scattering parameters over time to discriminate the healthy tissues from the cancerous cells. The stochastic data will be computed to obtain the overall signal characteristics over time. The precision of the diagnostic of the classifiers will be comparatively evaluated in each case. Ultimately the project will develop a low cost, wearable and safe breast cancer screening system for early detection and monitoring of breast cancer. |