Interpretable Machine Learning for Tumour Detection in Microwave Breast Imaging Using Image-Extracted Features

Recent advancements in machine learning and microwave imaging techniques have shown significant potential in automating tumour detection. Three classification algorithms were trained using features extracted from radar-based breast images. A logistic regression classifier demonstrated the best detection accuracy. Radar images were reconstructed using conventional delay-and-sum (DAS) and delay-multiply-and-sum (DMAS) beamformers, as well as an optimization-based radar reconstruction technique employing varying levels of enhanced physics modelling. Our results show that incorporating enhanced physics modelling improves tumour classification accuracy. This work applied a robust methodology that ensures diversity and separation between training and testing sets through the use of proper machine learning methods. By utilizing appropriate techniques, we can more accurately assess the algorithm's real-world applications.