Radiology genAI unveiled

Harrison.ai has recently introduced a radiology-specific vision language model. 

Called Harrison.rad.1, the dialogue-based model can perform open-ended chat related to X-ray images. It can detect and localise radiological findings and generate reports, all with an emphasis on clinical safety and accuracy. 

Unlike existing generative AI models trained on general and open-source data, Harrison.ai’s model has been trained on a set of real-world, diverse, and proprietary data that is annotated at scale by a team of medical specialists. 

The company further claims its genAI model outperformed major LLMs, such as OpenAI’s GPT-4o and Microsoft LLaVA-Med, in the competitive Royal College of Radiologists’ 2B Rapids examination, scoring an average of 50.88 out of 60 – on par with experienced radiologists. 

Additionally, Harrison.rad.1 showed 82% accuracy in the VQA-Rad benchmark dataset of clinical X-ray questions and answers and 73% accuracy in Harrison.ai’s in-house open-source RadBench dataset.

See-Mode bags 510(k) for thyroid ultrasound analysis AI

Victoria-based See-Mode Technologies has obtained the first 510(k) clearance from the United States Food and Drug Administration for an AI-powered solution to detect and diagnose thyroid issues in ultrasound scans. 

Its thyroid ultrasound analysis software detects single or multiple nodules and automatically classifies each based on the American College of Radiology’s TI-RADS rating systems. It also automatically generates a complete worksheet, which radiologists can immediately review and amend, and provides preliminary impressions following a clinician’s review and approval. Additionally, it streamlines the reporting of follow-up thyroid studies. 

New medical AI centre launched in Victoria

A new centre for innovations in medical AI has been established at La Trobe University in Victoria.

Backed by A$10 million ($6.8 million) in funding from the state government, the Australian Centre for Artificial Intelligence in Medical Innovation aims to develop AI to discover new treatments, vaccines, and immunotherapies for cancers, diseases, and viruses. 

An AI-powered colour map for tracking and predicting breast cancer spread and a biosensor for cancer cell detection are among projects in the pipeline. 

A media release also noted that the centre will be the first to access Nvidia’s supercomputer DGXH200 in Australia. 

National healthcare AI implementation guide sought in NZ

New Zealand general practices seek official national guidance on AI implementation and use in healthcare, a recent survey noted.

A group of primary care organisations, AI in Primary Care, surveyed GPs over the past two months to inquire about how AI is being used in primary care in the country. 

It was revealed that about a quarter of around 300 respondents are using AI daily in their work. 

Meanwhile, according to Dr Janine Bycroft, founder and CEO of Health Navigator Charitable Trust, the survey findings also point to organisations seeking “some sort of endorsement or framework from our national authorities that could be a game changer in terms of trust and acceptance for people.”

Currently, a comprehensive AI implementation guide by the WellSouth Primary Health Network is available to GPs in the country. Holding webinars and conducting privacy impact assessments are other ways primary care organisations have been promoting AI adoption within the sector.