Harrison.ai Radiology consolidates the former Annalise.ai product line under a unified brand, with vendor-reported deployment figures spanning more than 1,000 sites and 40-plus countries. The portfolio includes regulated software as a medical device for specific imaging tasks, an open platform for integrating partner algorithms, and a foundation model that the vendor explicitly describes as not cleared for clinical use. This profile distinguishes those categories because their regulatory status, intended use, and implementation requirements differ materially. [1][3][5][8]

Regulated Triage Products and Documented Findings

The vendor's Comprehensive Care solution is marketed for chest X-ray and non-contrast head CT triage, with claims that it identifies 13 findings in under 90 seconds and surfaces emergent and incidental findings for review. The marketing page lists 13 FDA-cleared findings including four intracranial hemorrhage types, obstructive hydrocephalus, acute infarct, vasogenic edema, pneumothorax, tension pneumothorax, pleural effusion, mass effect, and pneumoperitoneum. The vendor states this is the only FDA-cleared triage solution differentiating pneumothorax and tension pneumothorax on chest X-ray. [3]

The broader radiology solutions pages describe separate products with different finding counts: chest X-ray with 124 findings, CT brain with 130 findings, and CT chest with 167 radiological features. A FAQ document confirms the suite comprises chest X-ray, CT brain for non-contrast head studies, and CT chest for thoracic examinations. The vendor attributes these counts to training on large hand-labelled datasets, citing more than 229,000 unique CT brain studies and more than 782,000 unique chest X-ray studies annotated by over 140 radiologists. [4][5]

Workflow Integration and Configuration

The vendor describes a 'no click' viewer that displays AI findings within existing RIS/PACS workflows, with a resizable overlay window and findings grouped by priority or other customisable criteria. Marketing materials state that inference supports up to three chest X-ray images per case, including lateral views, and that the triage tool can flag vertebral compression fractures on chest X-ray. [3][4]

Configuration options are documented in the FAQ: customers can adjust which findings are enabled, how findings are grouped, and the detection thresholds applied to each finding. The vendor states that up to five different thresholds per finding can be configured, including within intracranial hemorrhage subtypes. The FAQ directs users to document cases where AI behaviour does not meet expectations and share them with a Customer Success Manager for clinical review and configuration recommendations. [3][5]

Implementation Boundaries and Unknowns

The sources do not specify hardware requirements, network dependencies, or integration timelines for the Harrison Orchestrator or the regulated triage products. The HOP page states that deployment conditions, technical specifications, and regional clearance affect availability, and discloses enterprise security certifications, data sovereignty with processing staying in region, OAuth 2.0/OpenID Connect authentication, hybrid cloud and on-prem orchestration, and compatibility with major PACS and RIS systems. The FAQ describes integration via standard DICOM and HL7 protocols. The FAQ also indicates that structured onboarding includes live product walkthroughs, workflow integration sessions, and recorded training materials, with post-go-live support through a Customer Success Manager and a support portal for tracking requests. [1][5]

A hypothetical demonstration could involve a radiology group evaluating the Comprehensive Care triage workflow on a de-identified non-contrast head CT dataset. The group might configure detection thresholds for intracranial hemorrhage subtypes, run the AI on a mixed-priority worklist, and compare the AI-prioritised ordering against their existing manual triage process. This scenario is illustrative only; the vendor sources do not describe such a pilot and no performance outcome should be assumed.

Harrison.ai Open Platform and Partner Ecosystem

The Harrison.ai Open Platform, branded HOP, is positioned as a single integration point for radiology AI from Harrison.ai and partner vendors. The vendor describes a Harrison Orchestrator deployed to the customer environment with one integration to PACS/RIS, and advertises zero platform markup with transparent per-study pricing. The stated rationale is to reduce vendor sprawl and IT integration costs by unifying multiple algorithms on one platform. [1]

The partner ecosystem listing includes vendors across modalities: AZmed and Radiobotics for X-ray, Therapixel and Lunit for mammography, Lucida Medical and AIRAmed for MRI, Nicolab for CT and MRI, Us2.ai and Koios Medical for ultrasound, CoLumbo for spine MRI, Nanox.AI and Sycai Medical for CT. The page notes regulatory clearances such as CE and FDA for individual partners, but the source does not provide a complete catalogue or specify which algorithms are available in which regions. [1]

Harrison.Rad 1.5 Foundation Model

Harrison.Rad 1.5 is a radiology foundation model that the vendor describes as capable of drafting reports from images using prior studies and clinical context. The technical report states it was trained on approximately 6.5 million studies yielding around 18 million conversations, with de-identified data following institutional and regulatory guidelines. The vendor describes a multi-stage training pipeline involving language model adaptation, vision-language contrastive learning, and instruction tuning over interleaved image and text inputs. [2][6]

The model is explicitly not cleared for clinical use. The technical report states it is released for research purposes only, is not a medical device, and must be used in accordance with terms and conditions. The product page repeats that no regulatory authority has reviewed, approved, or cleared the model for medical or clinical use, and that it must not be used as a substitute for professional clinical judgment. The vendor reports benchmark results on mock FRCR 2B examinations, including a median short case score of 86.5 against a pass mark of 73.2, while noting that actual exam questions are not publicly available. [2][6]

Regulatory Status and Organisational Context

The FAQ states that Harrison.ai products are regulated Software as a Medical Device with clearances and registrations in multiple markets, including TGA in Australia, FDA in the United States, and CE marking in Europe. It cautions that regulatory status varies by product and market and directs customers to their Customer Success Manager for current clearance information. The Comprehensive Care page separately notes that product availability and regulatory status vary by country. [3][5]

The October 2025 brand announcement states that Annalise.ai solutions are now known as Harrison.ai Radiology, with the transition not affecting existing solutions, agreements, or support. The vendor reports deployment across more than 1,000 sites, availability in over 40 countries, use by more than 3,400 clinicians, and nearly 7 million patient cases supported. These are vendor-reported figures from a corporate announcement rather than independently audited metrics. [8]