The future of medicine depends on patients who don’t exist

The most important problem that training different AI models presents is inaccuracies in the results. These inaccuracies or deviations can occur for many reasons. We may encounter algorithm biases, sample selection biases, feedback biases, or a lack of understanding of the social and/or cultural contexts of the selected models. But perhaps most important is that it is generated using initial data that is already biased and this occurs when the training data is not representative. Therefore, the way to avoid this is to use data as large as possible and as diverse as the reality analyzed. This is where the idea of ​​Generative Adversarial Networks (GAN) arises.

Adversarial generative neural networks (GAN) are characterized by the creation of mathematical vectors that have a discriminant that guarantees their effectiveness. They are a highly innovative deep learning architecture that has attracted a lot of attention in recent times due to its ability to generate highly realistic synthetic data that is difficult to distinguish from real data. Just as they have been widely used to date for the generation of images, 3D models or human faces, the production of artificial data practically identical to real data is a very interesting way of development.

The key to development is found in the identification, evaluation and classification of all possible variables in a clinical case and, subsequently, in the creation of the largest number of possible variations from them. In this way, the risk of bias in the training of AI is reduced because the way each system is trained can depend on the interests of each owner.

Obtaining data that can be trusted requires work in several directions:

A) The data gap from which the information is obtained is reliable and provides a high level of reliability in terms of the data obtained.

b) Quality protocols should be established that validate the said processes.

c) Validation mechanisms should be established by independent clinical and scientific teams that validate the data sets obtained.

To achieve this, it is necessary to have suitable IT partners that promote and enable these types of projects, such as the example of Alhambra IT.

Alhambra IT, through Alhambra Health (its specialized area in health), strives tirelessly to become a benchmark for health data quality, investing in various quality certifications (ISO 27701, ISO 27018, ISO 27001), In the security of both data and infrastructure, thanks to a solvent cybersecurity department that guarantees good management and administration of processes and clinical data.

Similarly, although it is often commented that data processing with artificial intelligence produces high CO impacts₂, This is an organization in Alhambra called ‘Neutral CO’₂That is, they eliminate or offset all emissions generated by their activity.

In particular, Alhambra IT has developed AI algorithms that allow creating virtual patients from real patterns. These patterns, which were previously unknown, are intended to train AI. The generation of these digital patients, also known as clinical clones or digital twins, allows them to be as numerous and diverse as needed by researchers, scholars and teaching staff in their respective activities, while complying with strict rules.

One of the relevant aspects when generating clinical cases is the difficult task on the part of clinical managers to select the most representative cases due to their frequency or lack thereof. This means that training professionals and preparing cases for their subsequent training involves heavy, slow and complex work that requires a very significant investment in hours in terms of selection, analysis and consensus by the clinical/scientific team. it occurs.

The use of clinical clones has provided professionals with a new way of working in both research and teaching.

We should not forget that till now doctors come trained but not trained. However, with the greater volume of cases due to digital twins, it becomes easier for professionals to improve their level of knowledge, not only in terms of the number but also in terms of the variants they can present.

There are some aspects worth highlighting about the use of clinical clones:

• This allows us to be more efficient when it comes to complying with data protection regulations (LOPDGDD).
• We can have a greater amount of clinical cases, both in number and variety of material, for professionals to study, to improve their level of knowledge and training.
• Artificial intelligence allows more information to be included in clinical episodes such as medical images, nurse notes, etc. As the professional progresses in his or her training, the inclusion of additional information allows the complexity of cases to progressively increase. In this way, artificial intelligence generates an ecosystem of cases that evolves based on the knowledge of each professional, while protecting the patient.
• Artificial intelligence provides data on professional practice by comparing humanization processes (greeting the virtual patient, interacting with him/her), requesting tests, and clinical practice guides as well as the application of vascular protocols.

In the last year, Alhambra IT has developed various projects with artificial intelligence based on clinical clones applied to hospital emergencies or different specialties such as dermatology.