Can artificial intelligence detect medical conditions mimicking psychiatric disorders sooner and prevent diagnostic delays

Introduction: Diagnostic errors are common and consequential, yet they don’t often receive enough research and policy attention. Several serious medical conditions mimic psychiatric disorders, resulting in substantial misdiagnosis and years’ long diagnostic delays. (1,2) Some examples of these conditions are multiple sclerosis, lupus, delirium, autoimmune encephalitis, and even brain tumors and leukemias. Symptoms of many medical mimics include mood changes, anxiety, depression, and fatigue, contributing to diagnostic complexity. Scarcity of time further adds to the diagnostic challenges. People who suffer from rare diseases (such as Wilson’s disease) encounter prolonged diagnostic delays, especially due to scarcity of experts. Diseases with multi-system symptoms too are on the receiving end of diagnostic delays and misattribution to somatic symptom disorder and anxiety. Diagnosis and getting to the cause of symptoms are core components of medical care, yet many people around the world suffer from prolonged untreated disease and complications. This can result in diminished trust in the healthcare system, with many patients reporting feeling dismissed, as shown by studies. Women are more commonly affected by autoimmune diseases and often experience years’ long diagnostic delay in detection of these conditions which are often initially attributed to psychiatric disorders. With artificial intelligence poised to shape healthcare, it is crucial to examine the role it could play in improving diagnostic detection.

Methods: A narrative review of Pubmed was conducted using search words such as “misdiagnosis”, “artificial intelligence”, “misattribution”, diagnostic delays; “autoimmune”, and others. Studies evaluating performance of artificial intelligence in complex medical diagnosis, diagnosis of autoimmune diseases, and other diseases that mimic psychiatric conditions, were included. Studies comparing physician and AI performance were also included.

Results: Studies show that large language models may outperform humans even in diagnosis of rate and complex diseases. (3) One study showed that in radiology and pathology, artificial intelligence enhanced accuracy and decreased diagnostic time by about 90% or greater. (4) Other studies show that artificial intelligence at least matches human performance in medical diagnostics. Artificial intelligence has also shown promise in increasing accuracy in detecting autoimmune diseases. This is in part because large language models can scan and interpret vast amounts of data rapidly. Studies show their benefit in detecting abnormalities on imaging in tumor detection too. Artificial intelligence can also help decrease administrative workload of physicians, allowing them to concentrate their attention on diagnostic detection. Large language models may exhibits lower thinking errors than physicians in diagnostics. However, ethical concerns, concerns around privacy of protected healthcare information, and challenges of standardization remain.

Conclusions: Diagnostic delays are a longstanding challenge in medicine, and generative artificial intelligence offers promise in enhancing diagnostic accuracy for medical conditions which commonly masquerade as psychiatric disorders. (5) Physician-experts need to be involved in research and sharing their real-world expertise with companies developing large language models to ensure adequate safeguards exist around the use of artificial intelligence in medical diagnostics.

References

1. Skegg K, Corwin PA, Skegg DC. How often is multiple sclerosis mistaken for a psychiatric disorder? Psychol Med. 1988 Aug;18(3):733-6. doi: 10.1017/s0033291700008436. PMID: 3186872. 2. Mitchell JL. Understanding the impact of delayed diagnosis and misdiagnosis of systemic lupus erythematosus (SLE). J Family Med Prim Care. 2024 Nov;13(11):4819-4823. doi: 10.4103/jfmpc.jfmpc_1177_24. Epub 2024 Nov 18. PMID: 39722963; PMCID: PMC11668484. 3. Abdullahi T, Singh R, Eickhoff C. Learning to Make Rare and Complex Diagnoses With Generative AI Assistance: Qualitative Study of Popular Large Language Models. JMIR Med Educ. 2024 Feb 13;10:e51391. doi: 10.2196/51391. PMID: 38349725; PMCID: PMC10900078. 4. Jeong J, Kim S, Pan L, Hwang D, Kim D, Choi J, Kwon Y, Yi P, Jeong J, Yoo SJ. Reducing the workload of medical diagnosis through artificial intelligence: A narrative review. Medicine (Baltimore). 2025 Feb 7;104(6):e41470. doi: 10.1097/MD.0000000000041470. PMID: 39928829; PMCID: PMC11813001. 5. Darrudi R, Hosseini A, Emami H, Roshanpoor A, Nahayati MA. Artificial intelligence in the diagnosis of multiple sclerosis using brain imaging modalities: A systematic review and meta-analysis of algorithms. Medicine (Baltimore). 2025 Sep 19;104(38):e44493. doi: 10.1097/MD.0000000000044493. PMID: 40988216; PMCID: PMC12459500.