Rare diseases represent a space of extremely high complexity and coexistence between the different medical specialties. That is why their diagnosis is very challenging and requires strict multidiscipline. Patients who suffer from these groups of pathologies must not only bear the burden of the disease itself, but also the overwhelming uncertainty of traveling a difficult path that often seems to have no end. Our job as healthcare professionals should always be to provide refuge for these people. To take care of them, to achieve some peace for them with the tools that we have at our disposal.
One of the tools that we have available today is artificial intelligence. This set of computational techniques allows us to make learning independent by having only one set of data in our hands. That is why today, especially in the current situation of the COVID-19 pandemic, we can transform those techniques into a beacon of light in the storm. In the following brief paragraphs, I will describe the opportunities that exist to use AI to empower the diagnosis of patients with rare diseases, regardless of where they are in the world.
Automatic analysis of medical images
Artificial vision algorithms are very versatile tools for generating automatic image classification.
In the specific field of medical imaging, a wide level of independence has been achieved with respect to medical professionals and in fact, in many cases the performance of algorithms has surpassed that of humans.
By combining a bio-image source and an algorithm that is adapted for its analysis, several strategies can be developed.
1. Using Ultrasound
Many rare pathologies, such as Fabry disease, develop myocardial hypertrophy and remodeling of the cardiac muscle. If an algorithm is trained with different ultrasound videos that represent progressive stages of cardiac involvement, after several iterations, it will learn to recognize the patterns that separate each stage.
2. Using X-rays and CTs
These imaging methods are especially useful for exploring lung involvement and organomegaly. For example, in pathologies such as lymphangioleiomyomatosis or pulmonary alveolar proteinosis where it is key to identify early lung involvement, an algorithm that automatically analyzes scans could be an excellent ally to generate an early diagnosis.
Digital pathology of biopsies
In many pathologies, the biopsy of the affected organs can not only confirm the diagnosis but also stage the patient. Knowing what the patient's state of compromise is, becomes crucial to start therapy and evaluate therapeutic progress with biomarkers. For example, in Fabry kidney biopsies, in Pompe muscle biopsies, and in localized scleroderma, skin biopsies are additional tools to determine clinical phenotype. The images resulting from these biopsies can be the inputs of a neural network that can identify not only the presence or absence of pathology but also different stages of involvement.
Automatic video surveillance of walking performance
Many pathologies that generate muscle degeneration (such as Pompe disease or Duchenne muscular dystrophy) can present a decline in walking performance as the muscular weakness of the limbs progresses.
Being able to analyze a yearly video sequence of a patient's walk is very important to understand the degree of progression of the pathology.
Therefore, with a high-definition video camera, an artificial vision algorithm can be trained. This would allow not only the recognition of multiple pathological stages but also the possibility of carrying out an active evaluation in the patient's home, avoiding the wear and tear of going to the hospital.
Real-time blood markers analysis
With the advent of automatic serial analyzers, which have 360-degree cameras for sample recognition, the analysis of blood and cerebrospinal fluid samples has accelerated exponentially. Combining this tool with classification algorithms such as the random forest or the support vector machines, it is possible to detect in real-time specific patients with some suspicious value of a rare disease. This allows the surveillance to be active, the screening of patients fully automatic and the optimization of hospital resources, the maximum.
Automatic analysis of Electronic Health Records
Thanks to natural language processing (NLP) techniques, it is now possible to automatically and remotely analyze thousands of health records in seconds. Automatic extraction of clinical markers can be performed and progression of markers over time can be evaluated. This analysis is crucial for the detection of rare diseases since algorithms can be trained to recognize unrelated clinical entities without a diagnosis. The algorithms could simply generate an alert to the family doctor or also directly generate an automatic order of evaluation by a specialist.
Self-assisted family linkage analysis
Another advantage of NLP and other AI techniques is that abstract maps of markers (blood, clinical and molecular) can be built and established as virtual units of comparison. In this way, an algorithm can automatically analyze if there are shared groups in families and suggest possible diagnoses. It should be remembered that the vast majority of rare diseases follow a Mendelian inheritance, which means that there are entire families affected. As clinical phenotypes can vary greatly with members living in very different geographic locations, implementing these units of comparison can be very important in reaching a precise diagnosis.
As we have seen, there are many opportunities to implement AI in the diagnosis of patients. The most important thing about AI is that once the algorithm is trained and tested with different data sources if it presents broad robustness, it can be considered an ideal candidate to assist the diagnosis. Moreover, all the "training" of the algorithm can be exported and shared with any computer in the world. This allows complex diagnostic techniques to be available in regions with little access to technology around the world. In fact, it can also be executed from mobile devices with a satellite internet connection, which would further enhance remote access.
We need to start considering that AI today can be an excellent ally in bringing hope and peace to many people who desperately need it. People who have perhaps been years without knowing what was wrong with their bodies.
Today more than ever, as we commemorate the day of rare diseases, our leadership needs to be aligned to improve the diagnostic methods and to ensure that everyone gets access to them. Each mom, each dad and each child.
Comments