Advanced machine learning and textural methods in monitoring cell death using quantitative ultrasound spectroscopy

A computer-aided-prognosis system is demonstrated based on quantitative ultrasound spectroscopy methods, textural features using local binary patterns (LBPs), and estimation of the distance between “pre-” and “post-treatment” samples using a kernel-based metric. The proposed method estimates the level of cell death, non-invasively, in preclinical animal models. In this study, sarcoma xenograft tumour-bearing mice were injected with microbubbles followed by ultrasound and X-ray radiation therapy successively as a new anti-vascular treatment. High frequency (central frequency 25 MHz) ultrasound imaging was performed before and 24 hours after treatment. Quantitative ultrasound spectral parametric maps were subsequently generated. Textural features were extracted by applying LBPs to the 2D parametric maps. Finally a supervised learning paradigm was used to classify the level of cell death to less/more than two threshold levels, i.e., 20% and 40%, by using the distance between the “pre-” and “post-treatment” images computed using a kernel-based metric. A maximum accuracy of 88% was achieved for both cell death level thresholds using the computer-aided-prognosis system and demonstrated the success of the system to identify the level of cell death, non-invasively in order to evaluate the effectiveness of cancer treatment administration.