Deep learning derived mechanical biomarkers for cancer therapy prediction

Objetivo

In the fight against cancer, it is well recognized that tumors are highly heterogeneous and they might differ considerably not only between tumors types but also among tumors of the same type or even for the same tumor during progression. As a result, the efficacy of standard cancer therapies varies, and while some patients respond to a particular treatment, other patients do not gain any benefit. Consequently, crucial in cancer therapy is the prediction of a patients response to treatment. Failure of standard therapies has led to the introduction of a new era of personalized, patient-specific treatments, which are based on the identification of biomarkers that characterize the state of a particular tumor. Many solid tumors (e.g. breast cancers and sarcomas) stiffen as they grow in a hosts normal tissue. Tumor stiffening is a known factor leading to compromised efficacy of therapeutics. Recently, it has been demonstrated by our team and co-workers that repurposing of common drugs with anti-fibrotic properties, known as mechanotherapeutics, target tumor stiffness and enhance therapy. Here, we aim to harness the power of deep learning (DL) methods in order to develop a robust biomarker based on ultrasound shear wave elastography (SWE). This biomarker will aim to: (i) predict patients response to treatment, separating responders and non-responders and (ii) monitor treatment outcomes, in the case of strategies that target tumor stiffness (i.e. mechanotherapeutics). The DL algorithms will be applied to a large set of existing preclinical data and to additional new data. A proof of concept clinical study on sarcoma patients will accommodate the clinical translation of the biomarker. Furthermore, we propose to develop a software product to be used as a commercial tool for the measurement of the DL-derived, SWE biomarker. A planned market research will highlight the best options for commercialization.