Artificial intelligence and disability: what progress has been made in improving medical research?
- Stéphane Guy
- Jan 17
- 10 min read
Artificial intelligence is revolutionising medical research on disability. By analysing millions of pieces of data (X-ray images, patient records, etc.) in a matter of seconds, it promises to diagnose conditions such as autism and multiple sclerosis more quickly, as well as predict their progression and identify new types of diseases. These advances could pave the way for more personalised and effective treatments. But what is the current state of research? We take a closer look.
In short
Studies show that AI is capable of detecting autism with 98.5% accuracy in children aged 2 to 4 years old by analyzing brain MRIs or simple retinal photographs.
The diagnosis of multiple sclerosis is accelerated thanks to algorithms capable of identifying new lesions using MRI scans.
New subtypes of diseases are being discovered by AI, paving the way for more tailored treatments for each patient.
AI could be able to predict the progression of disabilities, helping doctors anticipate patients' future needs and adapt care accordingly.
However, research faces challenges: the need for more data, the necessity for interdisciplinary collaboration, and the risk of bias in algorithms.
How is AI transforming medical diagnosis?
The case of autism: can it be detected before the age of two?
The diagnosis of autism spectrum disorders (ASD) is traditionally based on behavioral observation. However, this method can be subjective and delayed. In fact, in the case of autism, the process of assessing this disorder "is based on a range of clinical arguments gathered in various situations by different professionals."* In addition, "the diagnosis can be made from the age of 2."* In other words, an ASD assessment relies on a multidisciplinary team and a set of elements that enable a diagnosis to be made.
*Santé.Gouv.Fr, Dépistage et diagnostic de l’autisme, Fédération française de psychiatrie
However, the earlier it is detected, the more effective treatment can be. The question we might ask ourselves is: can autism be detected even earlier? ASD can be detected before the age of 2 by healthcare professionals. But AI could make it possible to shorten this time even further or enable more effective detection by analyzing medical imaging data with unprecedented accuracy.
In fact, an artificial intelligence system presented at the 2023 Radiological Society of North America (RSNA) conference analyzed imaging markers of autism in brain diffusion tensor MRIs. This type of MRI allows for non-invasive observation of the brain in order to detect (among other things) potential brain damage. It is a crucial tool for neuroscience and can be used to detect cases of autism, as in this example.
This AI-powered system diagnosed children aged 24 to 48 months with autism with an accuracy rate of 98.5%. Mohamed Khudri, a researcher at the University of Louisville in Kentucky, explains: "Our algorithm is trained to identify areas of deviation to diagnose whether someone is autistic or neurotypical."
*Eurekalert : Novel AI system could diagnose autism much earlier
Unlike current tests, which, according to sources, allow for reliable diagnosis from the age of 2, this imaging analysis method would not only lower the minimum age for diagnosing autism, but would also improve the accuracy of the diagnosis, with even more reliable results.
Could a simple photo of the eye be enough to detect autism using AI?
Researchers have even developed an even more accessible and less invasive approach: detecting autism through simple retinal photography. The results are promising. Artificial intelligence was able to accurately detect who was affected and who was not based on a simple photograph of the retina. The authors claim that "this AI model of retinal photography may be a viable candidate for an objective method of early screening for ASD, and possibly for the severity of symptoms."*
*Pourquoi Docteur : TSA : une IA détecte l'autisme chez les enfants grâce à la rétine
This method has several advantages: it is non-invasive, fast, inexpensive, and could be deployed in conventional medical practices, making early screening accessible to as many people as possible.
The case of multiple sclerosis: towards faster and more accurate diagnosis
Multiple sclerosis (MS) is an autoimmune disease that attacks the central nervous system. Early diagnosis is crucial to slowing its progression and improving patients' quality of life, but analyzing and detecting symptoms is a time-consuming process that requires specialized expertise.
How does AI overcome the lack of data to detect this disease?
Reda Abdellah-Kamraoui, a doctoral student at the Bordeaux Computer Science Research Laboratory (LaBRI), took part in the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) challenge on the detection of new lesions caused by multiple sclerosis. He explains: "[...] The problem is that since patients are treated as soon as lesions are detected, subsequent MRIs will not show significant differences, and we therefore lack data to train our algorithms. We therefore proposed a technique where we generate fake MRI images that simulate the case of a patient who has not been treated for several years, and then use them to train our AI." *
*Le Journal CNRS : Quand l’IA s’attaque à la sclérose en plaques
This approach helps get around a big problem in medical research: not having enough data to train algorithms, especially for conditions where patients get treated quickly.
Discovering new subtypes of multiple sclerosis using AI
More recently, scientists have identified two previously unknown biological subtypes of multiple sclerosis (MS) using artificial intelligence. Using a machine learning model called SuStaIn, researchers combined data from a blood protein (called sNfL) linked to the disease with brain imaging tests. This method revealed two subtypes of MS: "early sNfL" and "late sNfL." This discovery could help doctors tailor treatments more precisely to each patient.*
*Euro News : L'IA révèle deux nouveaux sous-types biologiques de la sclérose en plaques
This ability of AI to identify patterns that are difficult for the human eye to detect paves the way for precision medicine that is more tailored to each individual, where each patient receives treatment that is tailored to their specific biological profile and can benefit from faster, more accurate care. As a result, it will be possible to offer tailored treatment throughout the patient's lifetime.
Can AI predict the progression of a disability?
Anticipating to provide better care with artificial intelligence
Beyond diagnosis, AI could predict the progression of a disability, enabling healthcare professionals to develop personalized treatment plans and anticipate patients' future needs. This predictive approach represents a paradigm shift: rather than reacting to symptoms, doctors could now take proactive measures.
Algorithms can assess risks in a wide variety of contexts. For example, they can identify students who are likely to develop reading difficulties by analyzing their academic performance and learning styles. In the professional sphere, AI can predict the risk of permanent disability following a workplace accident by cross-referencing medical, demographic, and professional data.*
*Linklusion : Intelligence artificielle et handicap : révolution inclusive ou machine à exclure ?
What are the concrete benefits for patients?
This predictive capability of AI enables medical teams to:
Implement preventive measures before symptoms worsen
Adapt treatments according to the likely progression of the disease
Plan the necessary resources (technical aids, accommodations, support)
Informing patients and their families to better prepare for the future
The limits of prediction by artificial intelligence
However, this predictive capability of AI also raises ethical questions: how should these predictions be communicated to patients? What psychological impact might the announcement of an unfavorable prognosis have? How can we avoid AI treating the disease in an overly "mechanical" way, seeing it only as a "problem" to be treated, when it is actually a human being in all their richness and complexity?
Medical research is actively working on these issues to ensure that artificial intelligence remains an ethical tool serving patients, rather than a source of anxiety or discrimination. Once again, artificial intelligence is at the heart of ethical and social issues, including in the medical sector.
What are the current challenges facing research?
The significant need for quality data
One of the main obstacles to the development of effective AI in the medical field remains access to high-quality data that is sufficient in quantity and representative of patient diversity and, by extension, of different scenarios. As illustrated by the example of multiple sclerosis, researchers sometimes have to resort to image generation techniques to create data rather than relying on existing sources.
Furthermore, medical data is particularly sensitive and its use is strictly regulated by the General Data Protection Regulation (GDPR). Striking a balance between privacy protection and scientific progress remains a key challenge.
Essential interdisciplinary collaboration
Another challenge in the use of AI in the medical field lies in the transfer of knowledge between humans and machines. This transfer requires methodological advances in data processing and use, as well as in-depth collaboration between experts from different professional fields. It can be assumed that healthcare professionals and AI experts will need to work together, if they are not already doing so.
This need for close collaboration between different fields of study represents a significant organizational challenge for the various organizations and institutions involved.
Algorithmic bias: a risk that should not be overlooked
Current research on AI and disability is overwhelmingly dominated by a validist and medical perspective, which perpetuates discriminatory biases: "Nearly 60% of the studies reviewed approach disability from a purely medical angle: it is perceived as an impairment, a problem to be corrected or managed. This approach completely ignores the social model of disability, which states that disability arises from the interaction between a person with disabilities and environmental and social barriers (a building without a ramp, an inaccessible website, hiring biases)."*
*Linklusion : Intelligence artificielle et handicap : révolution inclusive ou machine à exclure ?
How can we ensure that algorithms are fair and respectful of people with disabilities?
To avoid algorithmic bias, several avenues are being explored:
Always include people with disabilities in research teams
Diversify training data to represent all populations and scenarios
Regularly audit algorithms to detect potential biases.
Adopting an approach to disability that integrates medical and social approaches
The involvement of people with disabilities: a decisive factor
The involvement of people with disabilities in the design and development of technologies capable of helping them is an important issue in order to develop tools that are truly adapted and capable of addressing the issues associated with the disabilities in question.
The risk of insufficient or inadequate consultation with people with disabilities is that tools will be designed that are unsuitable and do not meet their real needs, or only partially meet them. This raises a major challenge: how can we ensure that research meets real needs, and how can we effectively consult with the people concerned? This question is also relevant in the case of AI-assisted medical research.
What advances can we expect?
Ever earlier and more accurate diagnoses
Researchers are working on AI systems capable of detecting abnormalities even earlier, sometimes before the first symptoms appear. In the field of autism, for example, studies are exploring the possibility of detecting markers in the first months of life.
For other diseases such as Alzheimer's, Parkinson's, or amyotrophic lateral sclerosis (ALS), AI could help identify patients at risk years before the onset of the disease, paving the way for preventive interventions.
Multimodal AI: combining multiple types of data
One promising trend is the development of multimodal AI, capable of simultaneously analyzing different types of data: medical imaging, blood tests, genetic data, medical history, as well as behavioral and environmental data. This holistic approach could enable even more accurate and personalized diagnoses.
Precision medicine: already a reality?
Thanks to AI, precision medicine is no longer a futuristic concept but an emerging reality. In the future, each patient could benefit from treatment optimized according to their genetic profile, the likely progression of their disease, and their response to treatment.
The discovery of new disease subtypes, such as in the case of multiple sclerosis, is already making it possible to adapt treatment protocols. This trend is expected to grow in the coming years.
Research tools that are becoming accessible to everyone
Open source initiatives are emerging to democratize access to medical AI tools. The goal is to enable research laboratories around the world, including those with limited resources, to develop their own solutions and contribute to the advancement of knowledge.
This democratization could significantly accelerate the pace of discovery and promote more inclusive research that takes into account the diversity of global populations.
A collective responsibility for an inclusive future
While artificial intelligence offers extraordinary prospects for improving the diagnosis and management of disabilities, its development must be accompanied by in-depth ethical consideration. Nicolas Sabouret, computer scientist and professor at Paris-Saclay, concludes that “while AI designed solely to help people with disabilities is imperfect, can make mistakes, and still needs to be developed, it brings progress by putting energy and skills to work to improve the lives of people in need.”*
*France 24 : Quand l’IA améliore le quotidien des personnes en situation de handicap
According to the World Health Organization, one billion people (more than 15% of the world's population) live with some form of disability.*
*Un.org : 5 choses à savoir sur les personnes handicapées
The potential of AI to improve their medical care is immense, but it will only be fully realized if research truly involves the people concerned and adopts an equitable and inclusive approach.
FAQ
1. Can AI really diagnose autism with 98.5% accuracy?
Scientific studies show that AI can achieve this level of accuracy in children aged 24 to 48 months, thanks to the analysis of specific brain MRIs. Some research is even exploring detection through simple retinal photography. However, these tools do not replace a comprehensive clinical diagnosis by specialists, but are aids to early detection that can speed up treatment.
2. How does AI predict the progression of a disease?
AI analyzes thousands of cases of patients with similar characteristics and identifies patterns of progression. By cross-referencing medical data (imaging, biological analyses), demographic data, and environmental data, it can establish probabilities of progression. These predictions help doctors anticipate the patient's future needs and adapt treatment preventively, but they remain probabilities and not certainties.
3. Is the medical data used to train AI protected?
Yes, the use of medical data is strictly regulated by the GDPR in Europe. Data must be anonymized, and patients must give their informed consent. Research laboratories are subject to regular audits and must comply with strict security protocols. However, the issue of data security remains a major challenge that requires constant vigilance.
4. Why is there talk of bias in research on AI and disability?
Nearly 60% of studies adopt a purely medical view of disability, considering it to be an impairment that needs to be corrected. This approach ignores the social model of disability, which recognizes that disability also arises from environmental and social barriers. Furthermore, the lack of diversity in training data and the absence of involvement of the people concerned in the design of tools can lead to inappropriate or discriminatory algorithms.
5. Is medical AI research accessible to countries with limited resources?
This is a major challenge. Currently, most research is concentrated in developed countries with large infrastructures and significant budgets. However, open source initiatives are emerging to democratize access to medical AI tools. The goal is to enable all laboratories, regardless of their location, to develop solutions tailored to their populations and contribute to the advancement of global knowledge.
Comments