Today it is gaining relevance, also, in the management of both patients and examinations in large teams, the scanner (CT) and resonance (MRI). In its early days it was called computer-aided diagnosis (CAD). computer aided diagnosis). Concepts such as sensitivity, specificity, false positives are parameters that try to ensure reliability in the results. As digitization has been asserting itself in the Radiology, the versatility that is possible in it has been the subject of different improvements and advances, and also of criticism. The perspective that the passage of time gives us allows us to already have a framework that explains the nature of digital space and its degree of contribution to medical imaging diagnosis.
The approaches proposed in this topic so far can be grouped into two possible versions of the application of artificial intelligence, according to the two ways in which these techniques are also articulated in other fields:
– in the first, we have already commented, the computer would help the health professional in its functions, in those areas in which the results of the computer were more or less equivalent to those of humans or its performance was clearly superior;
– in the second, which was the first generation CAD strategy, it was regarding showing that the radiologist was expendable and the algorithm would replace the doctor. Thus, the following dichotomy might be proposed: Should we understand digitization as an aid to improve health services or as another reflection of the rise of machines in the hierarchy?
Who is the sovereign, the algorithm or the doctor?
We might deny the importance of the issue, but the truth is that we are not talking regarding a trivial issue. Who is the sovereign, the algorithm or the doctor? If the propaganda surrounding these techniques were just a decoration or a simple tactic of the big commercial houses, the matter would not be so important. But a sensation takes shape, and it is that the artificial intelligence can become a brake choice of certain medical specialtiesespecially in those in which new technologies play a more preponderant role.
It must also be taken into account that there are already many publications that announce that the role of algorithms can become very important. Many studies put the percentage of successes in detecting certain signs and radiological findings at more than 90%, ranging from breast canceremphysema, and even examples of detection of covid-19 pneumonia, with disparate results in the latter case. It must be taken into account that the errors do not have to coincide with the failures and the successes of the radiologists, so both results might be complementary.
|
“No digital algorithm is going to replace the debate of ideas in relation to making decisions regarding a patient, although it can make it easier for us to improve specific details” |
To elucidate these issues we should address at least the following. The first refers to the ability of algorithms to solve general questions, beyond specific findings. From the outset, the digital space is expressed as a one-way domain. Artificial intelligence and its algorithms work with a logic that has little to do with a multitasking system. That they are systems that train themselves and refine their algorithms by analyzing large amounts of data (the so-called deep learning), does not mean that they analyze all the pathologies of an individual at the same time, but by studying the characteristics that some individuals share with others, algorithms can be successful in the analysis of a given question.
As has been seen in multiple studies and articles, current AI systems, even the most advanced ones, respond relatively well to specific situations, but they have great difficulties coping with general problems. In other words, detecting a finding that follows a pattern in which a tumor with certain characteristics fits is not the same as recognize and diagnose a patient. So while it’s true that artificial intelligence can free us from many repetitive tasks, it’s not going to replace professionals, at least not in the foreseeable future. In addition, it is important to note that many of the news that appear in the media are in the research phase and have not passed the most basic phases of validation. But we must also take into account, yes, as Daniel Innerarity says, that examining groups of populations and creating profiles, as is the case, offers wide possibilities for exclusion and manipulation, and, we would add, for classification.
Understand the nature of algorithms in Medicine
The second section to understand the nature of algorithms, with the purpose of lowering them from their current enthronement, is the following: might digital logic render superfluous the way in which algorithm parameters are structured today? (self)sanitary organization of collaboration and debate to make diagnostic and therapeutic decisions, among health personnel in a health center, of collaboration between professionals of the different services in a hospital, of collaboration in the different committees, tumors, quality, etc.?
It does not seem possible in the foreseeable future, as we say, that algorithms can replace professionals. In the routine clinical workflow, homework cannot be reduced to a negotiation of digital algorithms. No automated digital scenario is in a position to forget the discussions and decisions that allow us to manage clinical day-to-day. This network of relationships is irreplaceable because, although artificial intelligence can resolve specific issues, fundamentally the most routine, it is not capable of leading the general system to quality level that would be required by current standards. Although specific action algorithms can be conceived, where we would be more limited by technology is in the resolution of conflicts and controversies due to different points of view, in many cases derived from small nuances and preferences, in the rationale for decision making to the most diverse questions.
But the question now is not whether the machines are going to replace all the professionals at once. Using a tone of complaint regarding what is said to seem inexorable and lamenting the changes that many professions are going to undergo in the more or less distant future can submerge us in an empty pessimism. Since the last decades of the last century, more than thirty years have passed and the expectations of the first CAD researchers were not met.
Ahead, important advances can be glimpsed (such as the so-called Radiomics; essentially a combination of AI in medical imaging with biological data y genetic), with fluctuations in intensity, which will depend on the ability of researchers and large commercial houses to manage the different elements at stake.
The outcome will be what it has to be, but artificial intelligence will have to submit to the ethical and democratic controls, will probably strengthen medical diagnosis and, in this way, hopefully lead projects that serve to improve people’s living conditions. In short, and regardless of its own limitations (remember that in multiple and varied fields, AI is in its infancy), no digital algorithm is going to replace the debate of ideas in relation to making decisions regarding a patient, although it can make it easier for us to improve specific details that can be derived from more routine needs.
