AI is helping clinicians spot rare diseases, uncover missed clues, and guide follow-up care in complex cases. New tools are flagging unusual autoimmune conditions, supporting radiology reviews, and improving eye and skin screening. Insights from medical experts show how these systems can sharpen diagnostic thinking while keeping clinical judgment at the center.
- Tool Flags Atypical Autoimmune Blistering Disease
- Anesthesiologists Assess Condition-Specific Perioperative Risks
- YEARS Deploys Assisted Eye and Skin Screens
- Timelines Reveal Overlooked Diagnostic Clues
- Reports Guide Faster Specialist Conversations
- Algorithms Prompt Targeted Radiology Follow-Up
- Clinicians Validate Expanded Rare-Disease Differentials
Tool Flags Atypical Autoimmune Blistering Disease
I had a case that presented initially as stubborn, treatment-resistant eczema on the hands and feet in a middle-aged patient, unresponsive to the usual topical steroid and barrier-repair approach over several months. Running the clinical photos through an AI-assisted differential tool flagged a pattern more consistent with a rare autoimmune blistering condition than atopic or contact dermatitis, based on subtle distribution and morphology cues that did not stand out to me on exam alone since the presentation was genuinely atypical for that diagnosis. That prompted a biopsy with direct immunofluorescence rather than another round of topical treatment, which confirmed the correct diagnosis and got the patient onto appropriate systemic therapy months earlier than the eczema-treatment cycle would have.
My advice for using AI on complex or rare presentations is to treat its output as a second differential opinion that earns a biopsy or referral, not as a diagnosis to act on directly. These tools are trained predominantly on common conditions, so their real value in a rare-disease scenario is flagging that something does not fit the expected pattern for your working diagnosis, which is exactly the moment a clinician should slow down rather than continue the current treatment plan. I would also caution against leaning on AI tools trained on limited or non-diverse image datasets for rare conditions that present differently across skin tones, since that gap is precisely where a confident-sounding wrong suggestion can do the most harm.
Dr. Cameron Rokhsar MD FAAD FAACS, Founder & Medical Director, New York Cosmetic Skin & Laser Surgery Center
Anesthesiologists Assess Condition-Specific Perioperative Risks
AI can be an excellent tool for situations where the patient has a rare condition. From the anesthesiologist’s perspective, it allows us to get a quick overview and “bird’s eye” perspective of the anesthesia-related considerations. For example, AI can quickly summarize how the condition affects airway anatomy, cardiovascular function, drug metabolism, or responses to specific anesthetic agents. Pulling all that information from medical journals or case reports can be time intensive and especially challenging if the surgery is urgent or emergent. My advice to others considering AI for complex cases would be to treat it as the starting point rather than the final answer. Clinical judgement and direct patient evaluation should be the primary driver for decision making.
Dr. Daniel Salazar MD, MS, D.ABA, Pediatric & Adult Anesthesiologist | Founder, EveryWipe™, Dr. Salazar Care
YEARS Deploys Assisted Eye and Skin Screens
At YEARS, we use AI-assisted funduscopy as well as AI-assisted full body skin screening.
We regularly use this for patients so I’d be happy to comment more if this is a fit.
Niko Hems, Longevity Expert. Head of Strategy & Medical Communications, YEARS
Timelines Reveal Overlooked Diagnostic Clues
In complex cases, AI does its best work as an organizer of information and a generator of questions. The pattern I keep coming back to is the patient with a decade of records scattered across several facilities, where nobody has ever put the sequence in order. When I load that history into a tool that builds a timeline, relationships surface that no single visit would reveal. Symptoms clustering after a specific exposure, or a finding noted once in an old report and never followed up on.
From there, I have it generate a differential, including the rare entries a busy clinic would reasonably deprioritize. What comes back is a list of possibilities worth investigating. Some of those suggestions are confidently wrong and read just as authoritative as the correct ones, so I test each item against what supports it and what argues against it.
Then the actual medicine happens. History taken in person, examination, targeted testing. Nothing on a machine-generated list becomes a diagnosis without that.
Two cautions I would give anyone doing this. Be deliberate about what identifiable data you put into which system. And be careful who sees the raw differential, because a long list of frightening rare diseases handed to a patient without clinical context creates a lot of anxiety for very little diagnostic gain.
Ben Frederick MD, Founder, Dr. Frederick’s Original
Reports Guide Faster Specialist Conversations
I entered my medical reports into an AI, where it became the first step to understanding them. Reading the documents wasn’t giving me any sense of what they were saying. The doctor confirmed and moved onto the next.
My advice for complex cases is to use it to comprehend your reports and to ask something in particular, and then take the answer to a specialist. It got me to the right conversation quicker. It didn’t end up becoming the doctor, and I wouldn’t want that.
Ankit Sarawagi, Curator, CFO Matrix
Algorithms Prompt Targeted Radiology Follow-Up
At Medicai, our radiology Co-pilot and specialized lung algorithm routinely act as companions to the radiologist, flagging atypical patterns during reads that prompt closer review and targeted follow-up. In practice this means the algorithm may raise a suspicion that leads the clinician to expand the differential and order additional imaging or consultation, which can uncover uncommon conditions. My advice is to treat AI as an adjunct, not a final arbiter: use it to surface findings and free clinicians from low-value tasks so they can focus on interpretation. Always confirm algorithm flags with clinical correlation and appropriate follow-up testing, and prefer specialized models that target the body part or condition at hand.
Andrei Blaj, Co-founder, Medicai
Clinicians Validate Expanded Rare-Disease Differentials
Can you describe a situation where AI helped you identify a rare condition?
I would describe it as AI providing a second set of eyes rather than making the diagnosis. In complex cases, AI can connect symptoms, longitudinal EHR data, imaging, genetic findings and medical literature that may be difficult to synthesize manually. Recent research is encouraging, a 2025 study found that an AI pipeline using EHR-derived phenotypes improved rare-disease differential diagnosis when combined with specialists. Another 2026 study demonstrated that LLM-assisted reanalysis of previously unsolved rare-disease genomes could surface new diagnostic leads.
What advice would you give to others about using AI for complex cases?
Use AI to expand the differential, challenge assumptions and identify overlooked connections, not to make the final clinical decision. Give it structured, high-quality information and ask it to explain the evidence behind each hypothesis. Most importantly, have a qualified clinician validate the findings against the patient’s complete history, examination and appropriate testing. Research also shows that AI performance can vary substantially in realistic clinical settings.
For me, the right model is simple, AI broadens the search, clinicians own the diagnosis. That balance can be especially valuable in rare diseases, where recognizing an unusual pattern is often the first step toward ending a patient’s diagnostic odyssey.
Noah Gula, AVP at OSP, OSP
