A drop of ink in a glass of water spreads quickly and evenly. But if the water is squeezed between tightly packed spheres, the ink cloud crawls slowly and erratically: it dodges obstacles and gets stuck in narrow necks. Water molecules in living tissue move in much the same way. In a brain tumor, cells are packed tightly, and the space between them resembles a labyrinth with constricted passages.
Conventional MRI shows this labyrinth in a blur. But by varying the time the scanner 'listens' to signals from water molecules, you can detect how far they manage to travel. At short delays, the water is just starting to spread and feels the nearest walls. At longer delays, its movement is hindered not by a single cell but by the whole tangle of pathways. This effect is called time-dependent diffusion.
The authors built the TRACED model, which decomposes the MRI signal into four components: the fraction of intracellular water, the spread of cell sizes, the true mobility of water in the intercellular gaps, and the tortuosity of these gaps (a parameter physicists call tortuosity). For fast signal computation, a neural network was trained on thousands of simulations of diffusion in virtual packings of spheres of varying diameters. To transfer this knowledge to real patient data, they used transfer learning: the network trained on simulations was fine-tuned directly during fitting to clinical measurements.
The model was tested on eight patients with gliomas of varying malignancy. In two cases, results were compared with histology — thin tumor slices examined under a microscope. Cell sizes obtained noninvasively matched the measurements on the slides. This means TRACED allows us to peer into tissue at the level of individual cells without making an incision.
This approach is reminiscent of a detective who, by the trembling of a spider web, deduces where the spider sits. Instead of a web, magnetic field gradients; instead of a spider, a cell. And the more accurately we read this trembling, the clearer we see the tumor's architecture.
Glioma cell sizes can vary tenfold — from 5 to 50 micrometers. That's like having a tennis ball and a watermelon in the same room.
In Philip K. Dick's novel 'Minority Report,' criminals are identified by the predictions of mutants. The TRACED model also 'predicts' tumor structure from indirect traces — only instead of clairvoyants, it uses magnetic field gradients.