Álvaro Yago Ruiz, Alexandra Prokhorova, Marta Cavagnaro, Marko Helbig, and Lorenzo Crocco.
An important and still open challenge in hyperthermia treatment is the non-invasive monitoring of the procedure, in order to verify the achievement of the therapeutic temperature in the target tissue, as well as to detect the presence of unwanted heating of healthy tissue, the so-called hot-spots. Given the known relationship between temperature and tissue electromagnetic properties, a deep learning based microwave imaging approach is proposed. Such approach employs the truncated singular value decomposition to provide an image which is then classified using a convolutional neural network. The case study of this work is neck cancer hyperthermia. In particular, two networks are used, one monitors the actual treatment of the tumour whereas the other monitors the presence of hot-spots in the spinal cord, a sensitive tissue to heat. To this end, the first network is trained to classify the tumor into one of three categories, {un-heated, therapeutic, too hot} while the second is trained to classify the spinal cord region into two categories {un-heated, too hot}. The proposed approach shows high performance accuracy.
