AI reaches the clinic: cardiac screening cleared, antibiotics discovered, rare diseases diagnosed
FDA clears cardiac AI that outperforms cardiologists on ECG; new antibiotics found for resistant gonorrhoea; rare-disease AI ends diagnostic odysseys.
By Dr. Asher Knippel
A week of regulatory milestones, published science, and commercial signals points to AI moving from research labs into routine clinical care — in cardiology, infectious disease, and rare-disease diagnosis all at once.
EchoNext Wins World's First FDA Clearance for Multi-Condition Cardiac AI
The US Food and Drug Administration has cleared EchoNext, developed by Pathway Labs, as the first AI system authorised to simultaneously screen for six distinct structural heart conditions from a standard 12-lead electrocardiogram. The conditions covered include left and right heart failure, valvular disease, severe hypertrophy, and pulmonary hypertension — a cluster of diagnoses that normally requires an echocardiogram, a more expensive and less widely available test.
EchoNext was trained on 700,000 paired ECG–echocardiogram records from NewYork-Presbyterian Hospital and validated on more than 500,000 patients across the United States and Canada. A Nature Medicine case study reported that the system detected 77% of structural heart problems compared with 64% by cardiologists reviewing the same dataset. Pathway Labs has announced a partnership with OpenEvidence to distribute the tool to clinicians.
The clinical significance lies in access. A 12-lead ECG is routinely taken in primary-care clinics, emergency departments, and rural health centres worldwide — including across Cyprus and the eastern Mediterranean, where echocardiography specialists are not always immediately available. A cleared AI layer on top of a standard ECG could shift detection of serious cardiac disease much earlier in the care pathway.
Monday, 22 June: AI Screens Six Million Compounds for New Gonorrhoea Antibiotics
Researchers at Karolinska Institutet and MIT published a study in Science Translational Medicine describing how a deep-learning model screened approximately six million chemical compounds to identify two antibiotic candidates active against Neisseria gonorrhoeae, the bacterium that causes gonorrhoea.
The lead compound, designated A1, kills both antibiotic-sensitive strains and multidrug-resistant strains. It works by targeting alanine racemase, an enzyme essential for bacterial cell-wall synthesis that has no counterpart in human cells — a mechanism that distinguishes it from currently approved antibiotics. The compound was also tested in an ex-vivo human-tissue model, where it reduced bacterial load measurably.
Gonorrhoea infects more than 80 million people each year globally. The World Health Organization has classified multidrug-resistant gonorrhoea as a high-priority pathogen, and resistance to ceftriaxone — the last reliably effective standard treatment — has been reported on multiple continents. These candidates are at a very early stage: neither has entered human clinical trials. Further development will require safety testing, dose optimisation, and eventually phase trials. Even so, a validated new mechanism of action against a pathogen with a fast-closing antibiotic window represents meaningful progress.
AI Diagnoses 18 Children With Rare Diseases After Years of Medical Uncertainty
A study published in NEJM AI by researchers at Boston Children's Hospital describes using OpenAI's o3 model to analyse the genomic and clinical data of 376 children with previously undiagnosed rare diseases — children who had often been evaluated for years without a confirmed diagnosis.
Working alongside human experts, the AI system yielded new diagnoses in 18 cases, a diagnostic rate of approximately 5%. The newly confirmed conditions included 10 neurodevelopmental disorders, four neuromuscular conditions, and two cases of early-childhood psychosis. For families, a confirmed genetic diagnosis is not merely a label: it can unlock access to targeted treatments, clinical trials, genetic counselling, and — often most importantly — an end to the exhausting process of searching for an explanation.
The study is a structured clinical evaluation, not a product launch. The use of a frontier language model for genomic interpretation raises important questions about validation, reproducibility, and clinical oversight that the field is actively working through. What the result demonstrates is that AI-assisted analysis of complex multi-modal data can surface diagnoses that expert human review of the same records had not reached.
Monday, 22 June: Insilico Medicine and SK Biopharmaceuticals Sign $2.5 Billion AI Drug-Discovery Deal
At the BIO 2026 conference in San Diego on 22 June, Insilico Medicine and SK Biopharmaceuticals announced a collaboration valued at up to $2.5 billion in development, regulatory, and commercial milestones, plus royalties. Insilico will apply its Pharma.AI generative-AI platform — which spans target identification, generative chemistry, and molecular optimisation — to discover drug candidates for neuroinflammatory, neurodegenerative, and rare neurological indications.
The headline figure warrants context: $2.5 billion represents the ceiling of contingent milestone payments tied to clinical and commercial success, not upfront capital. This is a preclinical-stage collaboration. Insilico has previously reported preclinical candidate nomination timelines of 12–18 months against an industry average of 2.5–4 years — a speed advantage that, if reproducible at scale, has significant implications for the cost and pace of drug development in neuroimmune disease.
CHOP Builds RareDAI to Steer Patients to the Right Genetic Test
Researchers at Children's Hospital of Philadelphia have published a description of RareDAI in npj Digital Medicine. RareDAI is a large-language-model-based clinical decision-support tool fine-tuned to analyse patient data and recommend the most appropriate genetic test for diagnosing rare diseases.
The problem it addresses is specific: rare-disease diagnosis requires choosing among hundreds of possible genetic tests, and the choice of test directly determines whether a diagnosis is reached. Inconsistent test selection across clinical teams contributes to the average diagnostic delay of five years or more that patients with rare diseases endure. Rare diseases collectively affect an estimated 300 million people worldwide; most are children, and most conditions are genetic in origin.
RareDAI aims to bring greater consistency and evidence-base to test selection, reducing the time and cost borne by families cycling through incorrect or incomplete testing. The tool is at the published-research stage and would require prospective clinical validation before routine deployment.
The content above is journalistic reporting and is not a substitute for professional medical advice, diagnosis, or treatment. Readers should consult a qualified clinician before making any change to their medications, treatment plan, or healthcare decisions.