Modality-Specific Cutoffs for 18F-Florbetaben Amyloid PET/CT and PET/MR: A Derivation and External Validation Study

In the modern understanding of Alzheimer's disease, the brain is no longer viewed solely through the lens of memory loss or confusion. Instead, doctors and scientists look for specific biological changes that occur years before symptoms appear. One of the most critical of these changes is the buildup of a sticky protein called amyloid-beta. When this protein clumps together in the brain, it forms plaques that are a hallmark of the disease. For a long time, seeing these plaques required examining brain tissue after death, but a breakthrough in medical imaging now allows doctors to see them in living patients. Using a special radioactive tracer injected into the bloodstream, a scanner can light up these plaques, turning an invisible biological process into a visible map. This capability has become essential not just for diagnosis, but for a new era of treatment. Powerful new medicines designed to clear these plaques are now available, but they are only safe and effective for people who actually have the amyloid buildup. Therefore, knowing exactly who has the plaques and who does not has shifted from a research curiosity to a life-or-death clinical decision.

However, seeing the plaques is only half the battle; measuring them accurately is the other. The images produced by these scanners are complex, and doctors must decide where to draw the line between a healthy brain and one with disease. This decision is often made by a trained human eye looking at the scan, a method known as visual assessment. But as treatments become more precise, relying on a simple "yes or no" glance is becoming less sufficient. Doctors need a way to measure the amount of plaque on a continuous scale, much like a thermometer measures temperature, rather than just guessing if it is hot or cold. This is where a standardized scale called the Centiloid comes in, which attempts to translate different types of scans into a common language of numbers. The challenge is that the machines used to take these pictures are not all the same. Some use a combination of a PET scanner and a CT scanner, while others use a PET scanner combined with a powerful MRI magnet. These different machines see the brain in slightly different ways, and a number that indicates disease on one machine might not mean the same thing on another. Without specific rules for each type of machine, a patient could receive a different diagnosis simply because they were scanned on a different device.

To solve this problem, a team of researchers from several major hospitals in China set out to create precise, machine-specific rules for a popular amyloid tracer called 18F-florbetaben. They gathered data from nearly seven hundred patients who had undergone these scans. The team carefully analyzed the images, comparing the computer-generated numbers against the expert opinions of radiologists who looked at the same pictures. They wanted to find the exact tipping point where a scan changes from showing a healthy brain to showing a diseased one. Crucially, they did not treat all the scanners as identical. They separated the data into two groups: those taken on PET/CT machines and those taken on PET/MR machines. They then calculated the specific number that best matched the experts' visual judgments for each group.

The results revealed a clear and significant difference between the two types of machines. For the PET/CT scanners, the researchers found that the tipping point for a positive diagnosis was a value of 15.1 on the Centiloid scale. For the PET/MR scanners, the tipping point was much higher, at 43.4. This means that a patient with a score of 30 would be considered to have the disease on a PET/CT machine, as it falls well above the specific threshold and the intermediate "gray zone" (11.4–26.3) established for that scanner type. In contrast, on a PET/MR machine, a score of 30 would fall within the gray zone, meaning the result is uncertain and not definitively positive. The study confirmed that using a single rule for both types of machines would lead to errors. By establishing these separate thresholds, the team ensured that the computer's measurement would align closely with what an expert doctor sees when looking at the image. When they tested these new rules on a separate group of patients from different hospitals, the rules performed with very high accuracy, correctly identifying the vast majority of cases of amyloid buildup and healthy brains, though a small number of cases showed minor discrepancies between the visual and quantitative readings.

The researchers also looked at the cases that were difficult to classify, the ones that fell right in the middle. In medicine, these are often called the "gray zone," where a doctor might hesitate to give a definitive answer. The study found that while the computer's gray zone and the doctor's uncertain cases often overlapped, they were not exactly the same. About a quarter to a third of the scans that doctors found difficult to read fell into the computer's gray zone, but many others did not. Furthermore, the study noted that when using a "double cutoff" approach to define this uncertain middle ground, a small percentage of patients (roughly 6% for PET/CT and 11% for PET/MR) were assigned to this gray zone, meaning they were not given a definitive positive or negative classification. This suggests that while the computer is an incredibly powerful tool, it does not replace the human eye entirely. Instead, the two methods complement each other. When a scan lands in the gray zone, it serves as a signal for doctors to look closer, perhaps re-examine the image quality or consider other medical information before making a final decision.

Ultimately, this work provides a practical guide for the era of amyloid-targeting therapies. By defining exactly where the line should be drawn for each specific type of scanner, the study removes a major source of confusion in clinical practice. It ensures that a patient in one hospital receives the same diagnosis as a patient in another, regardless of the machine they use. The findings offer a clear framework for integrating precise measurements with expert judgment, helping to ensure that the right patients receive the right treatments. As the field moves forward, these specific numbers will serve as the standard for interpreting these vital images, turning complex data into clear, actionable medical decisions.