User Avatar

I write about Crohn's disease and ulcerative colitis (IBD), digital health, diet, genetics and the microbiome. I am a passionate educator and innovator. I am a gastroenterologist and I run a busy IBD clinic. I am a Professor of Gastroenterology running a research team improving outcomes for people living with IBD. For more content please see charlielees.com and search Charlie Lees at YouTube.

How A.I. enhanced colonoscopy and pathology will disrupt IBD care, even where resources are limited

The application of artificial intelligence to routinely collected high-definition colonoscopy, capsule and digitised pathology images may herald the biggest paradigm shift in IBD care for a decade.

Automated generation of existing disease activity scores

Clinicians will have access to highly reliable machine generated endoscopic (e.g. Mayo) and histologic (e.g. Nancy) scores.

Application to (commercial) clinical trials could replace central reading which is time consuming and expensive (prohibitively so for anyone without the deep pockets of big pharma). This will accelerate development of new therapies.

Development of next generation scoring indices

These will provide a more thorough description of the characteristics, severity and extent of affected bowel segments. It will enable longitudinal tracking of patients over time no matter the disease distribution.

Novel A.I. predictive algorithms

This is where it gets really exciting.

Think about all the data we capture at colonoscopy, and what tiny fraction we actually base our clinical decisions on.

A.I. has the potential to unlock all the prognostic data that has so far been hidden within the noise. It may not work, but it is surely worth trying.

A stepwise route to development, testing & implementation

  • Build large banks of high-definition colonoscopy and digitised pathology

  • Annotate key features and train A.I. to generate disease scores

  • Build new disease scores based on additional feature sets

  • Test and retest to validate; iterate as more data generated

  • Implement in the clinic - software updates with minimal / no new hardware

  • Build longitudinal cohorts to train the AI to predict key outcomes - treatment response and flare - on images

How will this transform care? Improved standards, better reporting, better tracking of disease activity over time, democratisation of access to clinical trials, prediction of outcomes to enable personalised medicine.