One Image, Four Readings: Checking a Thermal Breast Result Against the Camera's Own Uncertainty

A thermal camera does not know a woman's skin exactly. To turn the camera's raw signal into a temperature, a few settings are fixed at one assumed value: how well the skin gives out heat (emissivity), the temperature of the room that reflects off the skin, the distance, and the humidity. The true values are never exactly these. They lie inside a small range. Today a thermal breast result is worked out once, from one image made with one set of values. This note proposes a different step. Make three or four versions of the same recording by moving the camera settings inside their range. Read every version with the classifier the centre already uses, and also work out one number from each version, the leading value of its temperature matrix. If all the versions tell the same story, the result is steady. If they do not, the result is fragile. A fixed rule then places the woman in one of three calm states, each with a defined action, printed by the computer from the numbers with no human step. No new hardware is needed and the classifier is not changed. This is an idea to be tested on real data, and not a result.