Why Harrison.ai thinks AI is the key to sustainable healthcare

Welcome back to Neural Notes, a column where I look at some of the most interesting AI news of the week. In this edition: Harrison.ai has just released its radiology-specific vision language model, the Harrison.rad.1.

I took the opportunity to catch up with the founders about the current state of play of AI and medtech, and what they’re planning next.

Last week Harrison.ai launched a new AI model specifically designed to assist radiologists examine x-rays and other imaging scans.

Named the Harrison.rad.1, it helps detect and pinpoint abnormalities in these images, generate detailed medical reports, and track a patient’s medical history over time to provide more accurate diagnoses.

The launch of Harrison.rad.1 is part of the company’s broader goal of easing the pressure on healthcare systems worldwide.

“Global healthcare is facing several challenges,” Harrison.ai co-founder and CEO Dr Aengus Tran told SmartCompany.

“There’s an increasing volume of imaging, a growing number of images that radiologists must review per case, a shortage of medical professionals, and a high psychological burden on the existing staff.”

For context, radiologists on average need to review thousands of images each week, often under time pressure. Harrison.rad.1 is designed to assist them by reducing the workload, allowing medical professionals to focus on more direct patient care.

According to Dr Tran, Harrison.rad.1 recently scored 51.4 out of 60 on the FRCR 2B Rapids exam, which is designed to test the skills of radiologists. This is a tough exam and only about 40-59% of radiologists pass it on their first try.

By scoring so high, Harrison.rad.1 has demonstrated that it can perform at the level of highly trained professionals and even outperformed other leading AI models.

Tackling burnout and the healthcare shortage

One of the main reasons Harrison.ai has invested in AI solutions is to tackle the ongoing issue of burnout in healthcare, a problem made worse by the COVID-19 pandemic.

Healthcare workers, including doctors and nurses, are facing mounting workloads. The risk of burnout is high.

Harrison.ai co-founder Dimitry Tran pointed out this is impacting the medical system both in and outside of cities in Australia.

“If you show up in the ED after 5 pm, you’ll be waiting five or six hours before you see a doctor in Sydney. Imagine rural Australia. Imagine the rest of the world. We’re very motivated to use AI as a way to give back time to doctors,” Tran said.

And this is a global problem. According to the World Health Organization, it projects a shortfall of 10 million healthcare workers worldwide by 2030.

“Doctors are just so tired, by the end of the day,” Dimitry Tran said, explaining how many healthcare professionals are forced to choose between spending time with their families and working through backlogs. He says AI can help alleviate this pressure.

“What computers can’t do is care for humans — hold someone’s hand and explain a cancer diagnosis, what options they have from treatment. We need to free them up from looking at black-and-white scans… to see patients. Not lock themselves behind computer screens. That is something a computer is good at,” Tran said.

Healthcare is a human right

Both founders see Harrison.ai’s mission as not just technological, but ethical.

“Healthcare is a human right. It’s not something only wealthy people, people with income should have access to,” Dimitry Tran said.

He also makes the point that AI makes it possible to scale medicine to help more people.

“This infinite capacity is something that has gotten us very excited,” Tran said.

This perspective is especially relevant for underserved areas where access to medical expertise is limited.

In many developing countries, and even in rural and regional Australia, people are often forced to travel long distances for diagnostics, delaying treatment and reducing the chance of recovery.

Harrison.ai aims to address this disparity by ensuring that healthcare providers in remote or underserved areas can access the same high-quality diagnostic tools as those in larger cities.

The company’s collaborations with providers like Sonic Healthcare and I-MED Radiology have been central to expanding the reach of its AI-powered solutions.

Turning AI ‘interns’ into ‘graduates’

Harrison.ai says its success lies in its ability to elevate AI models beyond basic, general-purpose tasks.

“I think of models like ChatGPT as great interns—capable of summarising an article or organising data — but you wouldn’t trust an intern to diagnose cancer,” Dimitry Tran said.

“What we can do is put it through medical school. We take a general-purpose model, like an intern, and give it millions of data points and hundreds of teachers to teach it so that it graduates into a specialist level, like a radiologist or pathology specialist,” Tran said.

This approach is reflected in Harrison.rad.1, which was trained using millions of medical images and reports, allowing it to perform complex diagnostic tasks in the area of radiology.

Training an AI to reach specialist-level expertise is not a straightforward process. General-purpose models like ChatGPT are excellent at handling wide-ranging tasks such as generating text or answering general questions, but when it comes to specialised fields like medicine, the margin for error is slim.

A misplaced diagnosis or a missed anomaly could have significant, even deadly, consequences for patient care. To address this, Harrison.ai focuses on a rigorous, data-driven approach where the AI is not only fed millions of data points but also fine-tuned under the guidance of domain-specific experts, including radiologists and pathologists.

This collaborative process ensures AI learns to recognise patterns and nuances in medical images that might be missed by less specialised models.

“We don’t just train the AI on any data—we train it on medical data specifically designed to mimic the work of radiologists and pathologists,” Tran said.

The final raise and plans for a multi-decade business

While the company’s current focus is on radiology and pathology, Harrison.ai has ambitious long-term goals. The company is looking to expand into other medical sectors, including oncology, cardiology, and ophthalmology.

“We see this as a multi-decade journey. We can keep doing more things and have more impact,” Dimitry Tran said.

Harrison.ai is also preparing for a Series C funding round, which it expects will be the last private raise before potentially going public. The company’s last funding round was for $129 million back in 2021.

The pair’s prior experience with their AI-powered embryo selection tool Ivy, which was the first of its kind to be commercialised globally, has shaped their approach to scaling Harrison.ai.

“Want to get this all the way to becoming an iconic listed company, because I believe that this is going to be multi decades. We will be using AI for many decades to come in healthcare, because that is the only solution,” Tran said.

Other AI news this week

Apple Intelligence officially debuted as part of the iPhone 16 launch.
Meta has admitted to scraping the public photos of every Australian on its platforms to train its AI. Unrelated, but can we also talk about Zuck’s glow-up, which is frankly concerning. I’m not sure I want to be in a Yassified Zuckerberg world. Bring back his aggressive sunscreen era, please.
While we’re on Meta — AI labels on Instagram and Facebook are about to be harder to find.
OpenAI Strawberry — a new model designed to fact-check itself and will be rolled into ChatGPT — has just been announced Hopefully, it will be able to spell its own name.
Speaking of OpenAI, unsurprisingly it is still trying for another raise. But what’s new is the valuation — the AI startup is aiming for US$6.5 billion at a US$150 billion pre-money valuation. For reference, its alleged valuation was sitting at US$86 billion earlier this year. That was bumped up to US$100 billion just two weeks ago. These are some eye-watering numbers, especially for a business that’s yet to turn a profit.
Adobe Firefly is getting a dedicated video generation model! Here’s a preview of it.