Application of Data Mining and Machine Learning in Microwave Radiometry (MWR)

Microwave radiometry has seen its way in successful usage in medical applications. The focus here is its applicability in cancer detection and monitoring, specifically for breast cancer, as an additional and alternative tool. This is done by capturing the temperature of the skin and the internal tissue. However, the amount of data required by clinical specialist to process in a short time to reach to a confident decision is becoming insurmountable. This can be tackled by developing a diagnostic system that will help pinpoint irregularities associated with pathologies. The key factors of a successful diagnostic system is the accuracy of the predictions and its informativeness and interpretability. The core component of such a system is a machine learning algorithm. Models that were explored were random forest, k-nearest neighbors, support vector machines, variants of cascade correlation neural networks, deep neural network and convolution neural network. From all these models evaluated, the best performing on the test set was the deep neural network. Also, we proposed a method for forming the space of thermometric features, which at the same time ensures a sufficiently high efficiency of the classification algorithms. More importantly, the model is inherently able to provide an explanation of the diagnostic solution.