Data Augmentation based on Inverse Transform Sampling for Improved Tissue Classification via Electrical Impedance Spectroscopy

Ultrasound-guided needle biopsy often leads to uncertain results due to poor visualization of targets in the image. Biopsy needles equipped with electrical impedance spectroscopy (EIS) sensors can be used to identify the tissue at the needle tip and provide more accurate needle guidance and lesion targeting. A machine learning algorithm is often used to classify the tissue based on the measured EIS spectrum. However, training machine learning algorithms requires large amounts of data, which is rarely available in biomedical applications. A solution to increase the size of the dataset and improve the performance of the classifier is to create synthetic data that closely mimics the original measured data. This paper proposes inverse transform sampling as a data augmentation method for EIS to bolster training dataset size. It exploits the cumulative distribution functions of the target data and uses inverse sampling to generate new, synthetic data. The method is demonstrated using an EIS dataset composing 13 different ex-vivo tissue types. The method is then validated by comparing the performance of the synthetic data to the original data through the use of an artificial neural network (ANN) and a convolutional neural network (CNN). The classification results indicate that classification sensitivity, precision, and accuracy increase by at least 27.38, 24.86, and 19.41%, respectively, when the classifiers used a mix of original data and data augmented with the proposed method.