Rare Disease and AI: Can AI help produce accurate and timely diagnosis?

Posted on: Wednesday 18 September 2024
Author: Mr Humphrey (Hung-Wei) Lin and Dr Tony Lockett

How to cite: Lin, H., and Lockett, T. (2024), 'Rare Disease and AI: Can AI help produce accurate and timely diagnosis?’, Faculty of Pharmaceutical Medicine, Wednesday 18th September 2024. Available at: [https://www.fpm.org.uk/blog/rare-disease-and-ai-can-ai-help-produce-accurate-and-timely-diagnosis/]

The article presents the views of the authors and not necessarily those of the Faculty of Pharmaceutical Medicine.

This blog post celebrates Jeans for Genes by exploring how AI advancements can help provide timely and accurate diagnoses for rare diseases like Frontotemporal Dementia, offering hope for early intervention and better care for those affected by genetic disorders.

Introduction

There are over 6,000 identified rare diseases [1]. A common theme among these diseases is the struggle that patients often have to get a diagnosis, the so-called Diagnostic Odessey. Early diagnosis not only brings relief, but early intervention is often beneficial. Recent advances in Artificial intelligence (AI) have led to the suggestion that AI could relieve the diagnostic odyssey. Frontotemporal dementia (FTD), a rare neurodegenerative disorder, exemplifies both the challenges and the solutions and this short article aims to explain and summarise these.

Like most rare diseases, a notable challenge for patients and their caregivers is the delay or errors in receiving an accurate diagnosis [2], which drastically restricts access to prompt medical or financial support. Despite initiatives like the Early Access to Medicine Scheme (EAMS), Innovative Medicines Funds, and the UK Rare Disease Framework, which aim to accelerate diagnosis and improve care, delays in accessing potential benefits still persist.

A study found that the average total diagnosis time (TDT) for people living with a rare disease (PLWRD) was 4.7 years, with a considerable amount of them waiting 9 months up to over 5 years for a confirmed diagnosis. Additionally, many of the patients consulted multiple healthcare professionals (HCPs) and many were misdiagnosed at least once [3].

As highlighted by the World Economic Forum and the WHO global report on AI in health research, advancements in AI and machine learning (ML) have shown promise in medical applications. These technology adoptions have already demonstrated potential in facilitating histology diagnoses in oncology [4,5] and differentiating Alzheimer’s disease from FTD [6].

FTD has a period prevalence of 0.16 – 31.04 per 1000 persons and accounts for 2.7% of all dementia cases and 10.2% in those aged under 65 [7]. The disorder presents many phenotypes, with bvFTD, svPPA, and nfvPPA being the most common clinically [8]. All phenotypes exhibit slightly different clinical symptoms affecting cognition, behaviour, executive functions, or language. While no known disease-modifying interventions yet exist, supportive pharmacotherapy and non-pharmacological management are available.

Early diagnosis and intervention are crucial for better prognoses in neurodegenerative diseases such as multiple sclerosis and FTD [9 –12]. However, FTD’s complexity and symptoms overlap with other neurological and psychiatric conditions making diagnosis challenging, especially when HCPs and caregivers lack disease education.

This suggests the potential for AI and ML to optimise the diagnostic process in rare diseases like FTD, offering patients a better chance at receiving the support they need in time.

Research article numbers rise with time

The variation in sample sizes and geographical origins in FTD studies suggest an imbalance in racial and ethical representations. This disparity may reflect ongoing challenges in clinical trial recruitment and delays or errors in FTD diagnosis. Most participants are from economically developed regions like Western Europe and North America. This suggests that FTD may still be underdiagnosed in parts of the world. The differences in study methodologies, expertise, resources, genetic risk factors, and cultural perceptions of behavioural dysfunctions likely contribute to this discrepancy [13].

Recent studies have increasingly relied on shared databases or registries like ALLFTD, FTLDNI, ReDLat, and GENFI to access larger sample sizes and more data to improve the statistical accuracy and credibility of diagnostic models. However, access to these platforms is uneven, particularly in developing or underdeveloped countries, where electronic clinical records systems may be limited, potentially leading to missed early signs of FTD.

Diversity in AI/ML models, data sources, diagnostic markers, and accuracy metrics

Various models, such as SVM, RF, CNN, and NLP, have been applied to explore FTD diagnostic classifications. The diversity highlights the versatility and adaptivity of these models and indicates positive further advancement. A slight modification to an existing model or applying model-stacking techniques can improve diagnostic prediction accuracy [14,15]. The increasing diversity of training data sources ranging from biomarkers to linguistic or behavioural cues also reflects the multidimensional nature of FTD and aids researchers in addressing the challenges of classifying multiple FTD phenotypes. Additionally, the wide range of diagnostic markers provides valuable resources for AI/ML models to interpret and generate accurate diagnostic results. However, the absence of consistency in the naming of the AI/ML models and the accuracy evaluation metrics has been noted, with a systematic review proposing the confusion metrics for more balanced reporting [16]. While most studies using AI/ML models yielded fair or good accuracy, the use of these varied terms and metrics complicates the objective comparison and evaluation.

Ethical and regulatory gaps

While studies involving real participants or database samples typically received ethical committee approval, the supervision might still be insufficient. Concerns about the AI/ML results validity as well as the safety and privacy of the data remain unclear and need further transparency. Interestingly, some studies indicated that patients and their families were less concerned about ethical and safety issues rather than having AI as the sole decision maker. They prioritised faster diagnosis and a balanced patient-physician relationship [17,18]. The European Union has taken steps to address these modern-day dilemmas with the EU AI Act, which sets limits on AI use and ensures that AI systems must not pose unacceptable risks to health, safety, or fundamental rights. Balancing these perspectives is essential for success in future AI medical integration.

Conclusion

The wide array of AI/ML diagnostic research angling from various aspects portrays strong interests, and the promising accuracy results propel AI towards effectively addressing this long-standing challenge with optimism and potential. Integrating these novel tools can enhance healthcare efficiency, giving patients and their families earlier access to essential care and improving quality of life. To support future research and optimise clinical implementation, consistency in terminology, a shared international registry, and clear AI-related regulatory guidelines are crucial. By addressing these issues, AI/ML diagnostic medical interventions could be successfully adopted into clinical practice and ensure early and precise FTD final diagnosis.

How can you get involved?

Join our Jeans for Genes fundraiser! Upload a photo in of yourself wearing denim (bonus points for double or even triple denim!), tag one of our social accounts and make a donation via our justgiving page.

Our social accounts:

Donate to our justgiving page to support Jeans for Genes

Donate
References:
  1. Nguengang Wakap S, Lambert DM, Olry A, Rodwell C, Gueydan C, Lanneau V, et al. Estimating cumulative point prevalence of rare diseases: analysis of the Orphanet database. European Journal of Human Genetics. 2020 Feb 1;28(2):165–73.
  2. Tsoukra P, Velakoulis D, Wibawa P, Malpas CB, Walterfang M, Evans A, et al. The Diagnostic Challenge of Young-Onset Dementia Syndromes and Primary Psychiatric Diseases: Results From a Retrospective 20-Year Cross-Sectional Study. Journal of Neuropsychiatry and Clinical Neurosciences. 2022 Jan 1;34(1):44–52.
  3. Faye F, Crocione C, Anido de Peña R, Bellagambi S, Escati Peñaloza L, Hunter A, et al. Time to diagnosis and determinants of diagnostic delays of people living with a rare disease: results of a Rare Barometer retrospective patient survey. European Journal of Human Genetics. 2024;
  4. Tian L, Wu J, Song W, Hong Q, Liu D, Ye F, et al. Precise and automated lung cancer cell classification using deep neural network with multiscale features and model distillation. Sci Rep. 2024 Dec 1;14(1).
  5. Guha A, Halder S, Shinde SH, Gawde J, Munnolli S, Talole S, et al. How does deep learning/machine learning perform in comparison to radiologists in distinguishing glioblastomas (or grade IV astrocytomas) from primary CNS lymphomas?: a meta-analysis and systematic review. Clin Radiol. 2024 Jun 1;79(6):460–72.
  6. Nguyen HD, Clément M, Planche V, Mansencal B, Coupé P. Deep grading for MRI-based differential diagnosis of Alzheimer’s disease and Frontotemporal dementia. Artif Intell Med. 2023 Oct 1;144.
  7. Hogan DB, Jetté N, Fiest KM, Roberts JI, Pearson D, Smith EE, et al. The prevalence and incidence of frontotemporal dementia: A systematic review. Vol. 43, Canadian Journal of Neurological Sciences. Cambridge University Press; 2016. p. S96–109.
  8. Tartaglia MC, Mackenzie IRA. Recent Advances in Frontotemporal Dementia. Vol. 50, Canadian Journal of Neurological Sciences. Cambridge University Press; 2023. p. 485–94.
  9. Singer BA, Feng J, Chiong-Rivero H. Early use of high-efficacy therapies in multiple sclerosis in the United States: benefits, barriers, and strategies for encouraging adoption. Vol. 271, Journal of Neurology. Springer Science and Business Media Deutschland GmbH; 2024. p. 3116–30.
  10. Freeman L, Longbrake EE, Coyle PK, Hendin B, Vollmer T. High-Efficacy Therapies for Treatment-Naïve Individuals with Relapsing–Remitting Multiple Sclerosis. Vol. 36, CNS Drugs. Adis; 2022. p. 1285–99.
  11. Wei W, Liu Y, Dai CL, Baazaoui N, Tung YC, Liu F, et al. Neurotrophic Treatment Initiated during Early Postnatal Development Prevents the Alzheimer-Like Behavior and Synaptic Dysfunction. Journal of Alzheimer’s Disease. 2021;82(2):631–46.
  12. Giovannoni G, Butzkueven H, Dhib-Jalbut S, Hobart J, Kobelt G, Pepper G, et al. Brain health: time matters in multiple sclerosis. Vol. 9, Multiple Sclerosis and Related Disorders. Elsevier B.V.; 2016. p. S5–48.
  13. Onyike CU, Shinagawa S, Ellajosyula R. Frontotemporal Dementia: A Cross-Cultural Perspective. In: Ghetti B, Buratti E, Boeve B, Rademakers R, editors. Frontotemporal Dementias: Emerging Milestones of the 21st Century. 1st ed. Springer, Cham; 2021. p. 141–50.
  14. Maito MA, Santamaría-García H, Moguilner S, Possin KL, Godoy ME, Avila-Funes JA, et al. Classification of Alzheimer’s disease and frontotemporal dementia using routine clinical and cognitive measures across multicentric underrepresented samples: A cross sectional observational study. The Lancet Regional Health – Americas. 2023 Jan 1;17.
  15. Moguilner S, Birba A, Fittipaldi S, Gonzalez-Campo C, Tagliazucchi E, Reyes P, et al. Multi-feature computational framework for combined signatures of dementia in underrepresented settings. J Neural Eng. 2022 Aug 1;19(4).
  16. Martin SA, Townend FJ, Barkhof F, Cole JH. Interpretable machine learning for dementia: A systematic review. Vol. 19, Alzheimer’s and Dementia. John Wiley and Sons Inc; 2023. p. 2135–49.
  17. Witkowski K, Okhai R, Neely SR. Public perceptions of artificial intelligence in healthcare: ethical concerns and opportunities for patient-centered care. BMC Med Ethics. 2024 Dec 1;25(1).
  18. Amann J, Vayena E, Ormond KE, Frey D, Madai VI, Blasimme A. Expectations and attitudes towards medical artificial intelligence: A qualitative study in the field of stroke. PLoS One. 2023 Jan 1;18(1 January).