{"slug": "october-2026-ai-ml-fda-clearances-a-close-reading-of-the-released-510-k", "title": "October 2026 AI/ML FDA Clearances: A Close Reading of the Released 510(k) Summaries", "summary": "FDA cleared 25 AI-enabled medical devices between September 26 and October 31, 2026, and had posted 510(k) summaries for 22 of them as of November 14, according to Innolitics' close reading of the released files. The review found that one reader study missed its primary endpoint, one device met its sensitivity criterion only on the point estimate, four devices cite a predicate built for a different organ or imaging modality, and FDA cleared four predetermined change control plans (PCCPs), including one whose summary names the PCCP as the only difference from the predicate. Among the clearances, Deep Reasoning AI's NeuroDR, developed with 3DR Labs, won FDA clearance on October 7, 2026, 104 days after receipt, on the strength of two standalone performance studies with no clinical testing, reporting expert accuracy scores of 4.65 to 4.93 on a five-point scale across 120 cases and average Dice scores of 0.95 (anterior circulation) and 0.96 (posterior circulation) across 50 cases.", "body_md": "I count 25 AI-enabled devices that FDA cleared from September 26 to October 31, 2026. As of November 14, FDA had posted the 510(k) summary for 22 of them, and this review draws on those summaries. Eighteen of the summary files are dated November 5 on FDA's server. It opens with three AI clearances for Innolitics clients, one from the window and two from earlier in 2026. Two of those devices went to FDA without a Pre-Submission meeting (Pre-Sub), and the third held one first. FDA cleared the three devices 104, 104 and 167 days after receipt. Among the other clearances, one reader study missed its primary endpoint, one device met its sensitivity criterion only on the point estimate, and four devices cite a predicate built for a different organ or imaging modality. FDA cleared four predetermined change control plans (PCCPs), including one whose summary names the PCCP as the only difference from the predicate.\n\nThe window starts where my [September review](https://innolitics.com/articles/september-2026-ai-ml-fda-clearances/) ended. I include a device when FDA Device Explorer, Innolitics' search tool for FDA records, flags it as AI-enabled or when its 510(k) summary or the company's announcement describes a machine learning function. FDA's [510(k) database file](https://www.fda.gov/medical-devices/510k-clearances/downloadable-510k-files) lists 327 decisions in the window. FDA had posted a document for 322 of them, and I searched those documents for machine learning terms. The files that hold 18 of the 22 summaries are dated November 5, 2026 on FDA's server, and the other four are dated October 2. As of November 14, FDA had posted no document for SwiftSight Body Composition. FDA's database entries for [QGuideCT](https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K260523) and [Ligence Heart](https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K261733) list a 510(k) statement in place of a summary, and the posted file for each holds only the clearance letter and indications. Review times count calendar days from FDA's receipt of the 510(k) to the decision. They include time on hold and exclude any Pre-Sub.\n\n[NeuroDR](https://fda.innolitics.com/device/K262134) takes a head and neck CT angiography (CTA) study and, with no manual steps, returns 3D and 2D views of the arteries that supply the brain. Deep Reasoning AI developed it with 3DR Labs. FDA cleared it on October 7, 2026, 104 days after receipt, with RapidAI's [Rapid Neuro3D](https://fda.innolitics.com/device/K243350) as predicate.\n\nThe summary reports two standalone performance studies and states that [no clinical testing was necessary](https://highlight.innolitics.com/v/rgfPwkN7kAci#h=1). In the first, clinical experts rated the clinical accuracy of the outputs of [120 cases](https://highlight.innolitics.com/v/rgfPwkN7kAci#h=2) against the source images, and scores ranged from 4.65 to 4.93 [on a five-point scale](https://highlight.innolitics.com/v/rgfPwkN7kAci#h=3). In the second, the software's segmentations of 50 cases had [average Dice scores of 0.95](https://highlight.innolitics.com/v/rgfPwkN7kAci#h=4) for the anterior circulation and [0.96](https://highlight.innolitics.com/v/rgfPwkN7kAci#h=5) for the posterior circulation against expert annotation. A Dice score of 1 is complete overlap.\n\nAccording to the [Innolitics case study](https://innolitics.com/portfolio/neurodr-neurovascular-cta-510k/), the team skipped the Pre-Sub, set the acceptance criteria from the predicate's published performance before the studies ran, and asked FDA for an early orientation meeting with a live demonstration. Work on the 510(k) began on February 2, 2026, the submission went to FDA 20 weeks later, and FDA cleared the device after one round of questions.\n\n[Vascular Assist Occlusion Triage](https://fda.innolitics.com/device/K260729) (VAOT) is triage software that flags CTA studies with a possible total occlusion of the aorta or of five named pelvic and leg arteries. FDA cleared it on June 18, 2026, 104 days after receipt.\n\nThe predicate is Aidoc's [BriefCase LVO Triage](https://fda.innolitics.com/device/K220709), which flags strokes on head CTA. The summary states that \"[The primary difference between the subject and predicate devices is the anatomical location of the suspected findings (Lower Extremities vs. Head).](https://highlight.innolitics.com/v/UVRH5bYSYKzh#h=1)\" On [317 cases](https://highlight.innolitics.com/v/UVRH5bYSYKzh#h=2) with ground truth from three radiologists by majority vote, [sensitivity was 91.5% and specificity was 90.1%](https://highlight.innolitics.com/v/UVRH5bYSYKzh#h=3) against an 80% performance goal for each. According to the [case study](https://innolitics.com/portfolio/radiology-cadt-in-9-months/), GuideAI skipped the Pre-Sub to keep its timeline. FDA raised two deficiencies during review and cleared the device nine months after project kickoff.\n\n[Heartvue.Proton](https://fda.innolitics.com/device/K260811) is a cloud-hosted cardiac MRI analysis platform with \"[25 machine learning models organized across 8 processing pipelines that together produce 48 quantitative cardiac measurements](https://highlight.innolitics.com/v/weD5zIOrcoa9#h=1)\". The interpreting clinician can review and edit every measurement. FDA cleared it on August 26, 2026, 167 days after receipt, on Heartvue's first submission. The predicate, Medis Suite MR-CT VVA ([K140587](https://fda.innolitics.com/device/K140587)), performs the same measurements manually. The primary endpoints measured error against reference values that at least three cardiologists agreed on unanimously, and the summary reports that [all of them met their criteria](https://highlight.innolitics.com/v/weD5zIOrcoa9#h=2).\n\nThe cleared PCCP has two modification types. \"MOD-1 (Retraining)\" covers retraining, and \"MOD-2 (Rearchitecting)\" covers changes to a model's architecture, \"[including consolidation of multiple models within a pipeline](https://highlight.innolitics.com/v/weD5zIOrcoa9#h=3)\". According to the [case study](https://innolitics.com/portfolio/heartvue-cardiac-mri-510k/), Heartvue held a Pre-Sub to get FDA's feedback on the predicate, the ground truth method, the per-model metrics and sample sizes, and the PCCP concept before funding validation studies.\n\n*Table 1. Schedule outcomes of the three client clearances. Days from receipt to clearance come from FDA's 510(k) database. Project durations and their starting events come from the linked case studies, and the starting events differ.*\n\n| Device | Pre-Sub meeting | Days from FDA receipt to clearance | Reported project duration | Starting event | \n|---|---|---|---|---|\n| [NeuroDR](https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K262134) | No | 104 | [247 days](https://innolitics.com/portfolio/neurodr-neurovascular-cta-510k/) | Start of 510(k) work on February 2, 2026 | \n| [VAOT](https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K260729) | No | 104 | [Nine months](https://innolitics.com/portfolio/radiology-cadt-in-9-months/) | Project kickoff | \n| [Heartvue.Proton](https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K260811) | Yes | 167 | [15 months](https://innolitics.com/portfolio/heartvue-cardiac-mri-510k/) | Start of the engagement in May 2025 | \n\nHeartvue's 15 months include the Pre-Sub. Its clearance is the only one of the three with a PCCP. The case study states that, for the changes the plan covers, Heartvue can retrain its models without filing a new 510(k).\n\nThe case studies give the reason for each choice. The NeuroDR team judged that comparable cleared devices had already answered the questions it would have asked, which were the predicate, the study design and the acceptance criteria. GuideAI had a predicate of the same kind and a timeline to keep. Heartvue's questions concerned the validation of 25 models and a PCCP, and the answers would set the size of its validation program. The NeuroDR case study says a Pre-Sub typically adds about three months before the 510(k) is filed.\n\nFDA also cleared a device for another Innolitics client in October, and it is not an AI device. Monebo Technologies' [Kinetic Software Library](https://fda.innolitics.com/device/K260640) (K260640, October 23) is an ECG analysis library that other manufacturers integrate into their devices.\n\n[NEXT-Cue](https://fda.innolitics.com/device/K261805) combines brain MRI, demographic data and cognitive test scores into a score from 0 to 100. The score supports a prognosis of progression from mild cognitive impairment to Alzheimer's dementia within five years. FDA cleared it on October 8, 2026, 129 days after receipt, with the BrainSee De Novo ([DEN220066](https://fda.innolitics.com/device/DEN220066)) as predicate. The evidence comes entirely from one research cohort, the Alzheimer's Disease Neuroimaging Initiative (ADNI), including validation on 180 participants from [12 US sites held out](https://highlight.innolitics.com/v/5kKHCoknpVyj#h=1) from model development.\n\nThe prespecified criterion was a sensitivity of 70%. The estimate of 71.3% met it on a point-estimate basis, but the lower bound of the 95% confidence interval was 61.6%, and the summary states that \"[these results do not demonstrate that sensitivity exceeds 70% with 95% confidence](https://highlight.innolitics.com/v/5kKHCoknpVyj#h=2)\". Specificity was 93.7% (95% CI 88.5% to 98.0%). The summary argues that the estimate is acceptable for a supplementary tool because patients with a false-negative output \"[are likely to receive continued clinical evaluation and longitudinal follow-up](https://highlight.innolitics.com/v/5kKHCoknpVyj#h=3)\". It also records that \"[The sponsor has committed to prospective non-ADNI U.S. post-market evidence generation for the locked released configuration.](https://highlight.innolitics.com/v/5kKHCoknpVyj#h=4)\"\n\n[CINA-Aneurysm](https://fda.innolitics.com/device/K262062) detects brain aneurysms larger than 3 mm on head CTA, marks and measures them, and sends a notification for triage. FDA cleared it on October 20, 2026, 124 days after receipt, with a PCCP. Tested without readers on 423 cases, the device met its criteria with an area under the ROC curve (AUC) of 0.954, a [sensitivity of 94.4% and a specificity of 91.7%](https://highlight.innolitics.com/v/66HDxOMgIfdJ#h=1).\n\nThe reader study missed its primary endpoint. Eight radiologists read 230 cases with and without the device, and their average AUC rose [from 0.9375 to 0.9497](https://highlight.innolitics.com/v/66HDxOMgIfdJ#h=2), a difference that was not statistically significant (p=0.075). AUC is a 0-to-1 measure of how well a reader separates positive from negative cases. The summary states that \"[Formal superiority was not established for the primary endpoint](https://highlight.innolitics.com/v/66HDxOMgIfdJ#h=3)\". It then states that \"[Per applicable FDA CADe guidance, clinical value was assessed on the totality of evidence across complementary endpoints.](https://highlight.innolitics.com/v/66HDxOMgIfdJ#h=4)\" Compared with unaided reading, sensitivity was 4.0 percentage points higher (95% CI 1.9 to 6.2) and reading time was 9.9 seconds shorter per case (p=0.0002).\n\nFour devices in the window cite a predicate built for a different organ or imaging modality, as VAOT did in June.\n\nTwo more devices differ from their predicates in method. Elucid's [FFR-IQ](https://fda.innolitics.com/device/K262015) (K262015, September 29) estimates fractional flow reserve (FFR), a measure of how much a coronary narrowing limits blood flow, from a CT scan with deep neural networks. Its predicate is the HeartFlow FFRct De Novo ([DEN130045](https://fda.innolitics.com/device/DEN130045)), and the summary states that \"[the predicate uses computational fluid dynamics and the subject device does not](https://highlight.innolitics.com/v/6wdQTi8uGOy8#h=1)\". On 491 vessels with invasive FFR as ground truth, sensitivity was 92.2% and specificity was 93.1%. The [EFAI NeuroSuite CTA LVO](https://fda.innolitics.com/device/K263109) (K263109, October 30), from Ever Fortune AI of Taiwan, is a deep learning stroke triage device that FDA cleared 56 days after receipt. Its summary describes the predicate, Rapid LVO ([K221248](https://fda.innolitics.com/device/K221248)), as using \"[traditional programming algorithms](https://highlight.innolitics.com/v/JBi7FxkD318M#h=1)\", and it cites a second cleared device for the AI method. Sensitivity was [96% and specificity was 94%](https://highlight.innolitics.com/v/JBi7FxkD318M#h=2) on 313 cases.\n\nSeveral summaries set the device's results beside the predicate's results from a separate study. Three of them follow.\n\n[Neurora](https://fda.innolitics.com/device/K260064) analyzes video and audio of a behavioral assessment to aid the diagnosis of autism spectrum disorder in children from 18 months through 12 years. FDA cleared it on October 2, 2026, 266 days after receipt, with the Cognoa De Novo ([DEN200069](https://fda.innolitics.com/device/DEN200069)) as predicate. In 173 children at five US sites, [sensitivity was 70.59%](https://highlight.innolitics.com/v/GtOL7TzxwNXy#h=1) and [specificity was 83.78%](https://highlight.innolitics.com/v/GtOL7TzxwNXy#h=2).\n\nThe acceptance criterion was a comparison with the predicate's performance across [its whole study population](https://highlight.innolitics.com/v/GtOL7TzxwNXy#h=3), which the summary puts at 51.6% sensitivity and 18.5% specificity. It derives those figures from the results table in FDA's De Novo summary with the predicate's No Result outputs included, and the predicate returned no result for 68% of its subjects. Among the subjects who did get a result, the predicate's sensitivity was 98.44%, and the summary states that Neurora's \"[Sensitivity is lower than the predicate's determinate-subset figure](https://highlight.innolitics.com/v/GtOL7TzxwNXy#h=4)\". Neurora returned a result for every subject in its analysis.\n\n[AiTiALVSD](https://fda.innolitics.com/device/K254302), from Medical AI Co. of Seoul, reads one 12-lead ECG and classifies patients as high or low risk for left ventricular systolic dysfunction, defined as an ejection fraction of 40% or less. FDA cleared it on September 30, 2026, 273 days after receipt, with Anumana's algorithm ([K250652](https://fda.innolitics.com/device/K250652)) as predicate. The model trained on [400,339 ECGs from 16 hospitals in Korea](https://highlight.innolitics.com/v/VEKHlINg4y6V#h=1) and was validated on 22,101 US patients. Sensitivity was 83.7% and specificity was 81.0% (80.4% to 81.5%). The summary lists the predicate's specificity as 83.6% (82.9% to 84.2%) and describes the difference in clinical performance as \"[not clinically significant](https://highlight.innolitics.com/v/VEKHlINg4y6V#h=2)\". The summary reports lower specificity in Black or African American participants and gives no subgroup estimate. It states that the study \"[was not powered to establish performance in this population separately](https://highlight.innolitics.com/v/VEKHlINg4y6V#h=3)\".\n\n[AVIEW Aorta Metrics](https://fda.innolitics.com/device/K260391) (October 16) segments the aorta on chest CT and measures its diameter. Its summary reports a [mean absolute error of at most 1.95 mm and a bias of -1.58 mm](https://highlight.innolitics.com/v/CBtGTJuYans8#h=1). Both figures are larger than the ones it cites for the predicate, Siemens' AI-Rad Companion ([K222360](https://fda.innolitics.com/device/K222360)), which are a mean absolute error of at most 1.8 mm and a bias within 1.5 mm. The confidence intervals of the two mean absolute errors overlap. The summary calls the result \"[comparable performance to the predicate device](https://highlight.innolitics.com/v/CBtGTJuYans8#h=2)\" and does not state the size of the test set.\n\nRIVANNA's BoneEnhance ([K262093](https://fda.innolitics.com/device/K262093), October 20) is a machine learning module that outlines bone and soft tissue on the company's Accuro XV ultrasound system. On [233 scans from 173 adults](https://highlight.innolitics.com/v/rykoW8GsL6sI#h=1) at five US sites, the average Dice score was 0.75, and the summary reports no clinical study.\n\nNaitive Technologies' OsteoSight Spine ([K261512](https://fda.innolitics.com/device/K261512), October 27) is [AI-enabled software](https://highlight.innolitics.com/v/C4mBNgFBWXFx#h=1) that reads spine and chest radiographs taken for other reasons in patients aged 50 and older. It notifies the physician when it estimates low bone mineral density. Its predicate is the Rho De Novo ([DEN230023](https://fda.innolitics.com/device/DEN230023)). Against bone density scans in [3,599 cases](https://highlight.innolitics.com/v/C4mBNgFBWXFx#h=2), sensitivity was 51% and specificity was 91%, and the summary reports [no prospective clinical study](https://highlight.innolitics.com/v/C4mBNgFBWXFx#h=3).\n\nNeurotronics' Sleep Analysis Module ([K261521](https://fda.innolitics.com/device/K261521), October 9) is a software library that estimates total sleep time from a photoplethysmography signal with [a locked machine learning model](https://highlight.innolitics.com/v/izx2fY2oyX5m#h=1). In [146 sleep studies](https://highlight.innolitics.com/v/izx2fY2oyX5m#h=2), its estimate differed from manual scoring by 3.02 minutes on average, with limits of agreement of -77.63 and +83.66 minutes. The summary notes that the acceptance criteria were [based on point estimates](https://highlight.innolitics.com/v/izx2fY2oyX5m#h=3) and that a difference of that size may be significant for an individual patient.\n\nFive other clearances reuse or extend existing AI functions, and none of their summaries reports a new clinical study. Cleerly's ISCHEMIA ([K262839](https://fda.innolitics.com/device/K262839), October 29) is a Special 510(k) for a software architecture change, and its summary states that [no changes have been made](https://highlight.innolitics.com/v/w8Ti1yEbUgPz#h=1) to the machine learning algorithms. United Imaging's uPMR Zenith PET/MR system ([K261676](https://fda.innolitics.com/device/K261676), October 28) includes deep learning functions cleared on earlier systems and [extends its bone segmentation to the whole body](https://highlight.innolitics.com/v/T67b7iYrkjBj#h=1), tested on [32 cases](https://highlight.innolitics.com/v/T67b7iYrkjBj#h=2). GE's Voluson Expert 18, 20 and 22 ultrasound systems ([K261933](https://fda.innolitics.com/device/K261933), October 2) [add AI software features](https://highlight.innolitics.com/v/gduqcfzdDXXk#h=1), among them SonoBiometry GYN and SonoGA, and the summary states that the submission [did not require clinical studies](https://highlight.innolitics.com/v/gduqcfzdDXXk#h=2). Brainlab's Spine Planning 3.0 ([K262257](https://fda.innolitics.com/device/K262257), September 30) adds planning features around AI algorithms that \"[were already cleared in K223553 and remain unchanged](https://highlight.innolitics.com/v/UN9HNTpFWPG9#h=1)\", and Cascination's CAS-One IR ([K261225](https://fda.innolitics.com/device/K261225), September 29) adds five structures to an existing segmentation function.\n\n*Table 2. AI clearances decided September 26 to October 31, 2026.*\n\n| Decided | Company | Device | K number | Product code | Review time (calendar days) | \n|---|---|---|---|---|---|\n| Oct 30 | Ever Fortune AI | EFAI NeuroSuite CTA LVO | [K263109](https://fda.innolitics.com/device/K263109) | QAS | 56 | \n| Oct 30 | Dentsply Sirona | DS Core Detect | [K261072](https://fda.innolitics.com/device/K261072) | MYN | 212 | \n| Oct 29 | Cleerly | Cleerly ISCHEMIA | [K262839](https://fda.innolitics.com/device/K262839) | QXZ | 79 | \n| Oct 28 | United Imaging | uPMR Zenith | [K261676](https://fda.innolitics.com/device/K261676) | OUO | 160 | \n| Oct 27 | Naitive Technologies | OsteoSight Spine | [K261512](https://fda.innolitics.com/device/K261512) | SAO | 173 | \n| Oct 23 | Deep Breathe | BlineSlide | [K261679](https://fda.innolitics.com/device/K261679) | POK | 155 | \n| Oct 20 | RIVANNA | Accuro XV with BoneEnhance | [K262093](https://fda.innolitics.com/device/K262093) | IYO | 120 | \n| Oct 20 | Avicenna | CINA-Aneurysm | [K262062](https://fda.innolitics.com/device/K262062) | QBS | 124 | \n| Oct 19 | Avicenna | CINA-Aneurysm+ | [K262147](https://fda.innolitics.com/device/K262147) | QIH | 116 | \n| Oct 16 | Coreline Soft | AVIEW Aorta Metrics | [K260391](https://fda.innolitics.com/device/K260391) | JAK | 252 | \n| Oct 9 | iSchemaView (RapidAI) | Rapid ICH v4.0 | [K262301](https://fda.innolitics.com/device/K262301) | QAS | 94 | \n| Oct 9 | Neurotronics | Sleep Analysis Module | [K261521](https://fda.innolitics.com/device/K261521) | OLZ | 155 | \n| Oct 8 | NeuroXT | NEXT-Cue | [K261805](https://fda.innolitics.com/device/K261805) | QWT | 129 | \n| Oct 7 | Deep Reasoning AI | NeuroDR | [K262134](https://fda.innolitics.com/device/K262134) | QIH | 104 | \n| Oct 7 | Ligence | Ligence Heart | [K261733](https://fda.innolitics.com/device/K261733) | QIH | 134 | \n| Oct 7 | Vista AI | Vista AI Scan | [K260092](https://fda.innolitics.com/device/K260092) | LNH | 268 | \n| Oct 5 | Medis Medical Imaging | QGuideCT | [K260523](https://fda.innolitics.com/device/K260523) | QIH | 230 | \n| Oct 2 | GE HealthCare | Lung Guidance | [K261042](https://fda.innolitics.com/device/K261042) | QJU | 185 | \n| Oct 2 | GE HealthCare | Voluson Expert 18, 20 and 22 | [K261933](https://fda.innolitics.com/device/K261933) | IYN | 115 | \n| Oct 2 | Argus Cognitive | Neurora | [K260064](https://fda.innolitics.com/device/K260064) | QPF | 266 | \n| Oct 1 | AIRS Medical | SwiftSight Body Composition | [K261612](https://fda.innolitics.com/device/K261612) | LNH | 139 | \n| Sep 30 | Brainlab | Spine Planning 3.0 | [K262257](https://fda.innolitics.com/device/K262257) | OLO | 90 | \n| Sep 30 | Medical AI | AiTiALVSD | [K254302](https://fda.innolitics.com/device/K254302) | QYE | 273 | \n| Sep 29 | Elucid Bioimaging | FFR-IQ | [K262015](https://fda.innolitics.com/device/K262015) | PJA | 106 | \n| Sep 29 | Cascination | CAS-One IR | [K261225](https://fda.innolitics.com/device/K261225) | JAK | 167 | \n\nThe 25 clearances took a median of 139 calendar days from FDA's receipt to the decision, with a range of 56 to 273.", "url": "https://wpnews.pro/news/october-2026-ai-ml-fda-clearances-a-close-reading-of-the-released-510-k", "canonical_source": "https://innolitics.com/articles/october-2026-ai-ml-fda-clearances/", "published_at": "2026-11-14 06:00:00+00:00", "updated_at": "2026-11-14 17:52:50.048138+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-policy", "machine-learning"], "entities": ["FDA", "Innolitics", "Deep Reasoning AI", "3DR Labs", "NeuroDR", "Vascular Assist Occlusion Triage", "RapidAI", "SwiftSight Body Composition"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/october-2026-ai-ml-fda-clearances-a-close-reading-of-the-released-510-k", "markdown": "https://wpnews.pro/news/october-2026-ai-ml-fda-clearances-a-close-reading-of-the-released-510-k.md", "text": "https://wpnews.pro/news/october-2026-ai-ml-fda-clearances-a-close-reading-of-the-released-510-k.txt", "jsonld": "https://wpnews.pro/news/october-2026-ai-ml-fda-clearances-a-close-reading-of-the-released-510-k.jsonld"}}