Last Month In Healthcare AI (September 2024)

Contents: ?? 1) Prediction & Diagnosis | ??2) Treatment & Care | ?? 3) Latest Approvals | ?? 4) Generative AI | ??5) AI Assistants | ??6) Governance & Ethics

?? 1) Prediction & Diagnosis

In cancer AI is being used for various types of prediction from detection, recurrence and associated chronic pain:

?? AI could help predict the risk of early-stage bowel cancer returning by analyzing immune cell levels in tumors. The AI-generated score identifies patients' risk levels, potentially reducing unnecessary chemotherapy and aiding doctors in making better treatment decisions

?? A deep learning model matched radiologists in detecting significant prostate cancer on MRI. When combined with radiologists, the AI tool improved diagnostic accuracy, potentially aiding in better biopsy decisions.

?? A new AI tool predicts which breast cancer patients are at risk of chronic pain by analyzing factors like anxiety and previous cancer diagnoses. This could help doctors personalize pain management for affected patients.

And examples from a wide array of therapy areas:

?? Researchers have developed an AI that claims 98% accuracy in diagnosing health conditions by analyzing tongue color, shape, and thickness. It could become a smartphone app for home diagnostics. The AI links specific tongue colors to conditions like diabetes, cancer, strokes, anemia, and COVID-19.

?? Researchers have developed an AI-powered blood test that can predict male infertility with 74% accuracy, potentially eliminating the need for initial semen testing.? This AI-enhanced approach could make infertility screening more accessible in primary care settings.

?? Google's AI model, HeAR, analyzes human sounds to detect health issues using a dataset of 300 million audio clips. It enhances the Swaasa? platform for early tuberculosis detection through cough analysis, aiming to reduce undiagnosed TB cases in India by making screening accessible and affordable

?? A new study shows that AI technology, using facial photographs from smartphones, can accurately detect common pediatric eye conditions like myopia, strabismus, and ptosis. This could enable parents to screen their children at home, allowing for earlier treatment and intervention.

?? Scientists will analyze 1.6 million brain scans using AI to develop tools that predict dementia risk. This initiative, part of the global NEURii project, will link CT and MRI images with health records to identify patterns, aiding early diagnosis and accelerating the development of more precise treatments for dementia

?? A deep learning model accurately predicted Clostridioides difficile infection (CDI) within 28 days of antibiotic treatment using electronic health records. Key predictors included body temperature and platelet count, potentially aiding in reducing CDI underdiagnosis and complications.

??2) Treatment & Care

AI is able to successfully triage patients, identifying those at highest risk:

??AI could help GPs identify high-risk heart patients, easing NHS pressures. The Optimise system, trained on over two million health records, identified 74% of those who later died from heart conditions. Early trials improved treatment, aiming for earlier interventions to reduce heart-related deaths and hospital admissions.

??Autonomous AI could significantly accelerate skin cancer diagnosis and reduce wait times. The research evaluated the AI tool DERM from Skin Analytics, used in skin cancer pathways, which showed a 99.8% accuracy in ruling out melanoma, surpassing traditional dermatologist evaluations

It is also showing promise in improving hospital operations and efficiency:

??Tampa General Hospital (TGH) has implemented Apella to enhance surgical operations. Apella provides real-time data on operating room activities, such as predictive case durations and staffing suggestions, helping surgical teams optimize workflows, reduce patient wait times, and improve outcomes.

??A study suggests AI could cut joint replacement surgery wait times by predicting the need for hip or knee replacements using natural language processing of radiology reports. Trained on 82 billion words of clinical text, the AI aims to improve efficiency and reduce costs.

??An NHS England-backed AI partnership with Deep Medical could save the local trust ￡28 million annually by cutting missed appointments by nearly a third. After a successful pilot, the AI solution is expanding to see 100,000-150,000 more patients, with plans to roll it out to ten more NHS trusts.

??3) Latest Approvals

?? Qure.ai's AI-powered chest CT solution, qCT LN Quant, has received FDA clearance. This tool aids radiologists and pulmonologists in analyzing lung nodules, tracking growth, and supporting lung cancer care.

??4) Generative AI

Gen AI continues to be used in the push to improve healthcare efficiency:

?? Elation Health has launched Note Assist, an AI-powered tool for their EHR platform, designed to streamline documentation during primary care visits. The tool transcribes and structures physician-patient interactions in real-time, reducing the documentation burden on clinicians.

?? Vivid Health's generative AI streamlines home healthcare by pre-filling 94% of paperwork, reducing staff workload, and doubling patient intake. It also helps create detailed care plans, addressing resource challenges and improving efficiency in home health services.

?? Kaiser Permanente is deploying Abridge's AI-powered note-taking tool across 40 hospitals and 600 offices, aiding 24,000 doctors. The tool streamlines documentation by converting patient conversations into structured notes, allowing doctors to focus more on patient care.

AI is being increaingly deployed in patient communication and education:

?? Researchers used ChatGPT to simplify echocardiogram reports, making them easier for patients to understand. A study found 84% of AI-generated explanations were accurate, with the rest mostly correct

?? ChatGPT could also save time in urology clinics by generating acceptable responses to 47% of patient questions, with better results on simpler queries

And when it comes to medical writing rapid advances are twinned with persistent limitations:

?? ChatGPT struggles with accuracy when medical questions are phrased in layman's terms. NIH researchers found AI tools work well with textbook descriptions but perform poorly with patient-written summaries

?? ChatGPT's medical writing has improved so much that it may soon fool AI detectors meant to spot non-human authorship. A recent analysis found that its manuscripts are increasingly indistinguishable from those written by humans, though issues like missing references and AI hallucinations persist

??5) AI Assistants

?? An AI assistant, Dora, developed by Ufonia, is helping ease pressure on the UK NHS by providing a faster, more efficient service for cataract care. Dora calls patients before and after surgery to assess their condition, identifying and prioritizing those needing further clinical attention. In the past year, it made over 12,000 calls

?? An AI-driven voice assistant, LOLA, effectively monitored heart patients after aortic valve replacements in the TeleTAVI trial. Among 274 patients, LOLA's follow-up calls helped identify complications early, enabling faster hospital discharges. Over half of the calls resulted in necessary medical interventions, with high patient satisfaction.

??6) Governance & Ethics

?? The WHO has released a toolkit to help countries assess their readiness to integrate AI into public health. It offers a framework for evaluating infrastructure, data management, and ethical considerations to ensure equitable AI implementation

?? Nearly half of FDA-approved AI medical devices lack clinical validation with real patient data, according to research published in Nature Medicine. The study analyzed over 500 AI devices and found that 43% had no published clinical validation data