Before AI reaches blue-light cystoscopy for bladder cancer — an observational study that registers the collection of training data itself (NCT07144319)

1Not a trial that evaluates an AI, but one that builds it

The phrase medical-AI clinical trial usually calls to mind measuring the performance of a finished system. This study sits one step earlier. Its purpose is to gather video and images from blue light cystoscopy performed in routine care, along with the matching clinical data, and assemble a training dataset. There is no algorithm yet, and so no performance to evaluate.

Medical AI depends first on data that is high in quality and diverse, and what marks this record is that the sourcing of that data has been registered as clinical research in its own right.

2Turning a subjective procedure into something measurable

The record states frankly that assessment during cystoscopy is subjective and carries large operator variability. Blue light makes lesions stand out, but the decisions about where to biopsy and what to remove remain with the human eye. The point of adding computer-aided detection is therefore not to win a contest over visibility; it is to narrow the spread in judgment between operators.

That the stated purposes extend to guiding how future validation should be designed suggests the regulatory path beyond data collection is already in view.

3Industry sponsors account for only 8 percent

Of the 750 medical-AI clinical trials this site holds as of 2026-08-28, 61 — 8.1 percent — are led by industry. Universities and hospitals lead 655 of them, or 87 percent, while government bodies and the National Institutes of Health account for 24 and 8 respectively. Clinical evaluation of medical AI is overwhelmingly academic, and a company starting from the collection of training data, as here, is in the minority.

What performance target the algorithm aims for, and when it would move to a validation stage, are not part of this registry record.

4Who is evaluating medical AI clinically

This study is a case of a company starting from the collection of training data. Splitting the trials this site holds by the class of the lead sponsor places it.

That is the breakdown across the 803 medical AI trials this site holds as of 2026-09-01. Clinical evaluation of medical AI proceeds overwhelmingly under academic leadership. The purpose here is to collect video, images and corresponding clinical data from blue-light cystoscopy performed in routine care and build a training dataset. There is as yet no algorithm and no performance to evaluate.