The Polytechnic University of Valencia (UPV) has promoted research carried out based on a doctoral thesis that aims to reduce “unnecessary” biopsies and allow the implementation of a low-cost prostate cancer diagnosis system in outpatient surgery centers.

The results of the work, which arise from the doctoral thesis of Juan Bautista Talens, coordinated by the professor of the Department of Electronic Engineering of the UPV José Pelegrí, have had the collaboration of the Research Institute for the Integrated Management of Coastal Zones of the UPV and the Research Institute of Hospital La Fe. Likewise, they have been published by BMC Medical Informatics and Decision Making, the UPV indicated in a statement.

The technique, which also makes the diagnostic process much less invasive, would in turn improve the method of classifying prostate cancer and reduce the number and cost of biopsies to eliminate its invasive nature.

“The intention is to devise a device that can be kept in a room and that, in a few minutes, tries to help doctors to be able to perform an analysis or better classify the disease,” explained Talens, while highlighting that “There is a problem and that is that many negative biopsies are performed.” “The technique itself to make it is carried out randomly,” she added.

Along these lines, he explained that “you try to extract tissue from the prostate and you are not sure that you are truly taking cancerous tissue, which generates many negative biopsies and, therefore, a high cost.” “In addition, they are invasive tests, because they have to enter through the rectum to remove the pieces of prostate,” she noted.

The solution derived from the research of Talens and Pelegrí – together with José Luis Ruiz and Tomás Sogorb – consists of extracting a series of signals through smell. “The passage of volatile compounds released from the urine is stimulated and passed through a chamber with gas sensors, from which a series of data is obtained from which an artificial intelligence (AI) is trained to capable of distinguishing cancer patients from those who have benign prostate hyperplasia, another disease that gives similar indicators,” he insisted.

“What we have done is use the electronic nose that we had, a device too large to take to a consultation to be able to identify and classify and, subsequently we have developed an entire engineering system around the software, the database and the electronic device; everything designed to be able to be taken to the consultation,” he continued, and stressed that “in the end a prototype, MOOSY4, has been achieved, which in a fairly advanced phase, could be useful in a clinical consultation.”

The model, which initially managed to correctly classify 87 percent of patient samples after indicating considerable effectiveness in its predictive ability, has been improved over time.

“We detected aspects to improve, especially in the detection of cases of prostate cancer with elevated prostate antigen (false negatives) to address this limitation. Therefore, we proposed a training strategy for the AI ​​that assigned greater weight to the cancer class compared to benign hyperplasia, and after implementing this adjustment, the results suggest an improvement in the model’s ability to handle more challenging cases,” Pelegrí added.

This research has several areas of future development, such as preparing the prototype to handle samples from patients with bladder cancer, improving the current electronic system using a SoC (System on a Chip), and continuing the development environment for future projects.