AI-assisted pathology tool tackles cancer coding

NEWS  - eHealthNews editor Rebecca McBeth

New Zealand's Cancer Control Agency Te Aho o Te Kahu, has successfully trialled an AI-assisted structured pathology reporting system that converts histology findings directly into fully coded structured data, and is preparing for its second phase.

John Fountain, data and analytics manager at Te Aho o Te Kahu, presented on ASAP (AI-assisted Structured Anatomical Pathology reporting) at the HiNZ Digital Health AI Summit where he said it significantly improves on the manual coding process that leaves the New Zealand Cancer Register months behind.

Pathology findings are traditionally sent from laboratories as unstructured text, which clinical coders must read, interpret, and manually translate into codes for registers and databases. 

This process is slow, resource-intensive, and replicated across multiple organisations that need structured cancer data, such as BreastScreen Aotearoa and the Breast Cancer Foundation NZ.

"Text is not data: we cannot use text directly for analytics, we cannot use text to auto-populate other systems that clinicians can interact with, and text basically must be read if you want to actually understand what is going on,” explained Fountain.

The trial, supported by Deloitte, AWS, and Health New Zealand, produced 117 fully coded data reports.
ASAP was developed using national interoperability standards - FHIR for data exchange and SNOMED CT for clinical terminology - to produce structured, coded outputs that are clinician-verified. 

It uses Te Aho o Te Kahu's national cancer data platform CanShare, which sits across a FHIR-based HealthLake platform within Health New Zealand and a Snowflake environment used for analytics.

The AI component involves AI scribe Tui, which converts the pathologists’ speech to text, then maps that text to SNOMED CT codes. The pathologist reviews the structured output before it is verified and sent through the system.

Coding accuracy during the trial was close to 100 percent said Fountain. However, the pathologists involved in the trial did not want to use the AI scribe as they preferred structured forms for them to follow.

"The takeaway message is you need to fit your solution to the clinical need, what the clinicians want, not replace the norm,” he told attendees.

The next phase, ASAP 2, will expand the scope to the full clinical data loop, from surgical request through to pathology report and into multidisciplinary meetings (MDMs).

Under this model, surgeons would submit coded requests to pathologists, pre-populated with relevant clinical information, and surgical notes would also feed into the system. The resulting structured data would then flow into MDMs, reducing the time needed for staff to prepare information for these meetings.

“Clinicians want instant gratification, so if their forms have been pre-auto-populated with information from the person that first collected data, it makes life easy,” Fountain said.

The Summit presentation was on September 17 in Christchurch.

Image: John Fountain, data and analytics manager at Te Aho o Te Kahu

If you would like to provide feedback on this news story, please contact the editor Rebecca McBeth.

You’ve read this article for free, but good journalism takes time and resource to produce. Please consider supporting eHealthNews by becoming a member of HiNZ, for just $17 a month.

Read more AI & Analytics news

Return to eHealthNews.nz home page