Colorectal cancer (CRC) is the third leading cause of cancer-related deaths when women and men are combined. The most likely screening detection technique is the endoscopy, but its reduced optical field of vision (<180º) can miss some polyps, for example due to angulations of the colon. To overcome this limitation, the host invented a low-cost hardware pre-prototype aiming to combine colonoscopy with microwave imaging (MWI). The physical basis of MWI is the dielectric contrast of the different tissues. These properties involve the relative permittivity and the conductivity, and their differences, are mainly related to water content. Malignant tissues have a higher rate of metabolism and more blood content, resulting in a higher dielectric contrast and electromagnetic (EM) scatter.
But, in addition to equipment, medical in-body MWI needs new software that allow highlighting of tissue differentiation and precise cancer detection. However, the foundations of the software require high skills in mathematics, computer science and physics. This project aimed to develop a reconstruction software able to detect CRC in a 360º field of view, and to provide techniques studying quantification, while reducing the computational burden.
The most successful MWI applications (breast cancer and stroke monitoring) locate many microwave antennas surrounding the imaging region at a specific distance and in a non-reflection medium. Instead, inside the colon, the number of antennas is restricted by the small organ’s sizes, the medium has reflections, the antenna’s location is unknown, and new geometrical configurations are involved. These changes restrict or inability the use of existing algorithms for other MWI reconstruction.
In this project, first a software based on a tomographic EM model was developed to simulate in-silico data. Then, the best developed detection software was tested over more than two thousand MWI in-silico datasets, to optimize the controllable parameters through a sensibility analysis. Finally, quantitative linear problems were developed, and an algorithm combining an iterative solver with an algorithm commonly used in compressed sensing (for problems dealing with few data) was adapted to reconstruct the quantitative MWI.
The detection software was able to detect the angular location of the CRC tissue, with a negligible angular error below 22.5º (smaller than the separation between the antennas). The quantification software showed its performance dealing with in-body scenarios such as the colon one, but for tissues with higher percentage (around 200%) of relative permittivity differences between malignant and healthy tissues. The project served to transfer software to a new spin-off aiming to commercialize the pre-prototype.
