Hypothesis

The foundation of this project is based on the following hypotheses:

1. • Digital transformation is driving a revolution across industries, but even more so in the electromedical sector, where equipment is increasingly expected to leverage the advantages of this transformation and the applications of artificial intelligence for the benefit of patients, as well as for greater business competitiveness.

   • Medical imaging equipment is likely one of the areas most positively impacted by digital transformation, as demonstrated by the numerous artificial intelligence (AI) applications in the most advanced imaging modalities, such as MRI and CT. However, few of these applications have achieved widespread clinical implementation.

   • Within the field of medical imaging equipment, so-called “conventional radiology” is perhaps the slowest in adopting these new paradigms. Patient workflows and the process of interpreting and reporting results remain essentially the same as in the analog era, except for some specific support tools that generally provide limited diagnostic assistance, typically within a restricted range of pathologies.

   • In conventional radiology, there is still considerable room for improvement in the use of AI tools during workflow stages prior to diagnostic interpretation by the radiologist. Many of these stages have a significant impact on the quality of the imaging result and, consequently, determine both the efficiency of the image acquisition process and the final image quality (always relative to the radiation dose received), as well as the need for repeat tests, which in turn increases the radiation dose.

   • One limitation of plain radiography compared with CT—whose use in all cases is unfeasible due to its cost and high radiation dose—is its lower sensitivity in detecting lung lesions. Tomosynthesis can substantially increase diagnostic accuracy in plain radiology systems, providing pseudo-tomographic information at a much lower dose. Incorporating AI techniques into tomosynthesis image reconstruction could overcome the main limitation of tomosynthesis, namely that it does not provide quantitative information. This would make it possible to generate images comparable to CT scans with a low-cost, more widely available device—important given the high number of pathologies associated with increased pulmonary parenchymal density.

   • SEDECAL, the industrial partner in this proposal and an international leader in the design and manufacture of human and veterinary imaging systems (particularly conventional radiology), has identified a significant niche opportunity in the digital transformation of these systems, through the incorporation of AI techniques in many different aspects. Seizing this opportunity will allow the company to consolidate its international leadership in the sector while, of course, improving patient care.

   • Gregorio Marañón Hospital and Universidad Carlos III maintain a multicenter research group that in recent years has achieved significant advances in the application of AI to various aspects of radiological image acquisition, with potential industrial interest and application. Likewise, the hospital has highly qualified professionals capable of defining clinical needs and validating interim results as they are achieved, though without extending to direct clinical application within the framework of this proposal.

   • SEDECAL and the academic partners of the proposing consortium have long collaborated, and with the support of the hospital’s Innovation Support Unit, they have achieved a high level of coordination and effective transfer of results in previous projects.

   • Consequently, the consortium believes there is a high probability that the prior scientific developments considered in this project can be successfully transferred, reaching approximately TRL-5, so that in the near future they may be fully validated within the same collaborative framework and incorporated into SEDECAL’s manufactured products. This would strengthen a public-private collaboration that has already proven to be a model of success in our environment.

Objectives

Assuming that the fulfillment of these hypotheses creates an ideal framework for successful academic–industrial collaboration, we can define the overall objective of the project as follows:

To design and develop a chest imaging system based on tomosynthesis and to complete the development of prior scientific results in the field of AI, implementing their necessary adaptation to the real industrial environment, so that the image acquisition workflow becomes less error-prone, provides higher-quality images with greater diagnostic accuracy, reduces the patient’s radiation dose, and ultimately enhances the company’s competitiveness while strengthening the public–private collaboration strategy between hospitals and industry.

This overall objective is broken down into the following specific objectives, which reflect the concrete contributions to the different stages of the radiology workflow, and which are subsequently addressed in the project activities:

• O1: Design and develop a new chest radiology system that includes hardware, software, and firmware with robotized movement of the X-ray source to enable tomosynthesis projections.

• O2: Incorporate AI tools to facilitate the preparation process prior to image acquisition by automating tasks such as: automatic detection of radiological positioning, automatic selection of voltage and current values, and automatic collimation. This will support the work of the radiology technician and avoid the need for test repetitions, thereby reducing radiation dose.

• O3: Incorporate AI tools in the image generation stages to enhance the information provided through: (1) improvements in contrast, noise, and detail definition, (2) reduction of scattered radiation without the need for an anti-scatter grid, and (3) elimination of tissue overlap, thus bringing the pseudo-tomographic image closer to that of lower-cost, more compact systems.

• O4: Apply AI to image interpretation, both as a first step in quality control—alerting to the need for test repetition—and as a second step in diagnostic support, by estimating the probability that the patient has pulmonary involvement based on the acquired images. Where applicable, this will include assigning a “score” and enabling visualization of the findings on the images.

• O5: Generate the databases necessary for training the algorithms and validate the results obtained in a relevant environment, which in our case will be a hospital room specifically equipped for testing. Hospital staff will carry out the independent tests required to ensure achievement of a TRL-5 development level, thereby allowing immediate clinical validation following this project. The priority clinical validation area will be lung cancer screening. (HGGM and UC3M).