Multimodal breast cancer risk classifier based on biomaterial impedance analysis

Purpose of the research. Breast cancer is the most common malignant tumor among women in Europe and its early detection plays a leading role in reducing mortality rates. Currently, X-ray mammography is the standard screening method for detecting breast cancer. However, due to the morphological similarities between benign and malignant lesions, many of the positive screening mammograms are false positive (up to 40%). Therefore, automation and intellectualization of this process is an urgent task. Methods. The presented studies examine the problems of finding new, highly sensitive, prompt and non-invasive methods for detecting malignant tumors, based on the use of modern computer and telecommunication technologies, which make it possible not only to identify early manifestations of a pathological focus, but also to monitor the process of the effectiveness of therapy without significant harm to the patient’s health. Results. The presented model of a multi-channel classifier integrates the capabilities of multi-frequency bioimpedance measurements and matrix acquisition of information from the surface of human skin through multi-electrode matrix systems. To do this, based on a matrix of electrodes, 3D mapping of the skin surface in problem areas is carried out. Through multi-frequency scanning, we obtain a three-dimensional bioimpedance image, which is analyzed by a convolutional neural network and/or by a decision maker. The proposed solution allows simultaneous analysis of data by an expert (bioimpedance image) and a convolutional neural network (trained classifier), which leads to a reduction in false positive results. Conclusion. The possibilities of multichannel monitoring open up prospects for constructing impedance multidimensional "portraits" of malignant tumors. To classify “portraits” (diagnostics and preclinical diagnostics), methods and algorithms for image recognition and classification are used.