CATEGORY 76 AUSTRALASIAN DENTIST ARTIFICIAL INTELLIGENCE We are asking humans to do too much: The limits of dental imaging By Dr Sen Le and Daniel Seo Dr Sen Le Daniel Seo In our first article, we made a simple argument: dentistry’s next leap is not digital, but intelligence. Digital changed how we capture information. Intelligence changes what we do with it. Our images have never been sharper, nor our workflows faster. Yet our diagnoses are no more accurate than before. This article is about why. The answer is uncomfortable. Modern imaging now holds far more information than the human eye was ever built to read. The image has outgrown the eye Modern imaging is staggering in its detail. A single bitewing typically holds two to four million pixels. A panoramic image can hold up to six million and every one of those pixels can encode more than 65,000 shades of grey. A cone-beam CT (CBCT) scan dwarfs them both with more than 1 billion voxels, each carrying thousands of shades of its own, up to 65,536 at the highest bit depths.1 The human eye can distinguish about 60 shades of grey.2 That is the gap. We now capture more than any eye can take in, then ask ourselves to find what matters inside it, often in seconds. A bitewing is read in under a minute.3 A CBCT that deserves twenty minutes may get less than five.4 And volume is only part of the difficulty. A flat radiograph collapses three-dimensional anatomy onto a single plane, so overlapping structures can hide pathology. The earliest and most treatable signs, such as incipient caries or subtle early bone loss, are faint, low in contrast, and easily lost.5,6 None of this is a failure of skill. It is the reality of human perception applied to a task that has quietly outgrown it. What the evidence shows The consequences are measurable. Reading radiographs unaided, dentists catch fewer than half of the early caries, the very lesions still small enough to reverse.5 Ask two skilled clinicians to assess the same images and they frequently disagree. In one reliability study, 14 dentists varied substantially in how they identified periapical lesions, and even the same clinician sometimes classified the same image differently when reviewing on a different occasion.7 This is not a question of skill. It is what happens to even the most expert eye, working at the very edge of what human perception allows. Periodontal disease: where the limits show Periodontal disease is the clearest example. Despite being largely a preventable disease, it affects up to 50% of the global population with 1 billion people affected by its severe form. Due to its gradual and silent nature, it is no surprise that the average time from onset of disease to diagnosis has been found to be 56 months. Detecting it early means comparing fine changes across a full series, and ideally against images taken months or years earlier. That is exactly the slow, comparative pattern the eye handles poorly. Key insight: Variability in radiographic interpretation is not a gap in competence. It is the predictable result of asking expert clinicians to detect subtle disease across dense, greyscale images, at speed, and to do so consistently, image after image, patient after patient. We have spent two decades making the image sharper. We have never asked whether the eye could keep up.
RkJQdWJsaXNoZXIy MTc3NDk3Mw==