The Algorithm Will See You Now: AI Reads the Radiograph

For a century, the dental radiograph has been read the same way: a clinician, a lightbox — later a screen — and a trained eye scanning greys for the darker grey that means trouble. That reading is now increasingly shared. Software cleared by the FDA for radiograph analysis sits inside a growing number of American operatories, outlining suspected caries, measuring bone levels, and flagging what a hurried human glance can miss.

What the machine actually does

Today’s dental AI is narrow by design. Systems cleared for market — a category pioneered by companies like Overjet and Pearl — are pattern detectors trained on millions of annotated radiographs. They draw boxes and margins: a suspected carious lesion here, a measured distance from cementoenamel junction to bone crest there. The dentist remains the decision-maker; the software is a highlighter, not an oracle.

That distinction matters clinically. Detection studies consistently show the value is in consistency, not genius. A tired human misses things at 4:40 on a Friday; the model reads image ten thousand exactly as it read image one. In periodontics, where millimetres of bone loss separate “watch” from “treat,” an instrument that measures the same way every time is genuinely useful.

What it cannot do

The model sees the film, not the patient. It does not know that the darkness on a root surface is an old, arrested lesion your dentist has watched for five years. It does not weigh your caries risk, your habits, your history — and it has no concept of whether treatment now serves you better than monitoring. Overtreatment is a human decision, and a highlighter in the wrong hands can become a sales tool: a screen full of red boxes is persuasive to a patient even when half those boxes merit nothing but a recall visit.

What it means for you in the chair

Three practical takeaways. First, if your dentist uses AI-assisted imaging, that is broadly a good sign — practices investing in current diagnostics tend to invest in current everything. Second, the boxes are conversation starters, not verdicts: ask “would you treat this today, or watch it?” and a good clinician will happily separate the urgent from the theoretical. Third, a second opinion remains what it always was — the strongest instrument patients have. An algorithm inside one practice does not replace an independent human outside it.

The technology is young, the trajectory is clear, and the winners will be the practices that use it the way it was cleared to be used: as a second set of eyes, with the first set still responsible for looking you in yours.