Analysis of Microwave Scans of Cancer Patients for Classification of Treated and Untreated Tissue

This paper examines the feasibility of detecting short-term changes in tissue related to cancer treatment. 15 female patients enrolled in a clinical trial, who underwent lumpectomy and accelerated partial breast radiotherapy, were recruited. The patients were scanned using a microwave imaging system at 2 time points - the first scan is at least 8 weeks after surgery but before radiotherapy began, and the second scan is at 6 weeks after the conclusion of radiotherapy. The permittivity or the dielectric property of breast tissue was estimated from the scan data and used for initial analysis. Next, wavelet transforms were applied to the time-domain response to extract features that were then classified. The Discrete Wavelet Transform (DWT) and the Dual Tree Complex Wavelet transform (DTCWT) were used to extract meaningful features that were evaluated using the Support Vector Machine (SVM), Random Forest (RF) and the Gradient Boosting classifier (GBC). After hyperparameter tuning of the classifiers, the experimental results show good accuracy, with SVM yielding the best results. In conclusion, the classification pipeline proposed for microwave imaging shows potential to track short-term changes in tissue as a result of cancer treatment.