#61–Rajpaul Attariwala, M.D., Ph.D: Cancer screening with full-body MRI & a masterclass in radiology

KEYPOINTS

• 01

  Radiation dose varies by orders of magnitude across imaging tests

  Attariwala puts a mammogram at roughly 0.05 millisieverts, which he calls negligible, while a whole-body PET-CT stacks the CT's x-ray dose on top of radioactive fluorine-18 glucose and can easily total 30–40 millisieverts. Attia notes the US regulatory ceiling is 50 millisieverts a year and that simply living at sea level contributes about 2–3, more at altitude or with radon. His rule: if you are going to irradiate someone, extract as much information as possible from that exposure.

• 02

  A CT is an x-ray done in three dimensions, calibrated so tissues are comparable anywhere

  CT uses Hounsfield units, with water fixed at zero and air and bone at the extremes. Because the eye cannot see the whole range at once, radiologists 'window' the image, centring it on lung, bone or soft tissue. Attariwala says that for non-moving organs such as the pancreas, slice count matters less than people think; 256-slice machines mainly help with fast-moving structures like the heart.

• 03

  Ultrasound is fast and radiation-free but depends heavily on the operator

  High-frequency sound waves bounce back from tissue boundaries, which makes ultrasound ideal for quick bedside answers and, through a window that avoids lung air, for watching heart valves move. Attariwala stresses that the skill of the person holding the probe can't be replaced, which is the main reason results vary so much.

• 04

  Mammography accuracy depends on breast density

  Attia recalled all-comer figures of about 84–85% sensitivity and 90–91% specificity, while acknowledging the numbers may be dated. Attariwala said averages hide the real story: in fatty breasts sensitivity can exceed 95%, but a single mammogram in dense glandular tissue may be only about 55% sensitive, because low-energy x-rays don't penetrate that tissue well. That is why some women are told they also need an ultrasound.

• 05

  Sensitivity and specificity measure two different failure modes

  Using 100 women with cancer, 80% sensitivity means 20 are missed (false negatives). Using 100 women without cancer, 90% specificity means 10 are wrongly flagged (false positives). Attia pushes listeners to ask which error a test makes, and in whom, rather than accepting a single accuracy number.

• 06

  MRI uses strong magnets to align hydrogen in water and fat

  Field strength is measured in tesla (one tesla is 10,000 gauss), and scanners commonly run at 1.5, 3 or 7 tesla. Higher fields align more hydrogen nuclei, which gives more signal. Contrast-free 'dry' MRI is usually described as harmless, but Attariwala described a patient with severe migraines whose headache was triggered by the scanner.

• 07

  Diffusion-weighted imaging finds cancer by how freely water moves

  DWI, or DWIBS with background suppression, compares water motion at two points within microseconds. Tumours are densely packed with cells, so water diffusion is restricted. Attariwala compares this to how a surgeon feels a hard lump. Pairing this functional signal with anatomical detail approaching 1×1×1 mm is his version of PET-CT's 'one plus one equals three', without ionising radiation.

• 08

  The main harm of screening is the false positive, not the scan itself

  Attia calls whole-body cancer screening a very personal decision that isn't for everyone. Outside of rare issues like a magnet-triggered migraine, the harm is what a false alarm sets off: anxiety, extra imaging, biopsies and the small but real risks of each procedure.

• 09

  MRI quality isn't standardised, so 'a DWI MRI' isn't one product

  Unlike Hounsfield-calibrated CT, MRI signal varies with vendor, coils and protocol. Attariwala calls it buyer beware. The Quantitative Imaging Biomarkers Alliance (QIBA) at the Radiological Society of North America is pushing standards organ by organ, starting with prostate and liver, with breast following. Attia admits that even he couldn't judge differences between scanners in another city.

• 10

  Machine learning helps most where the problem is narrow

  Algorithms are already being tested as a second reader for mammograms, especially for comparing a woman's current images with her previous ones. Attia floats 97–98% sensitivity and specificity as the level needed for confidence. Attariwala thinks that will be hard to reach because breast compression and positioning differ each time. In whole-body images, even telling which organ is the liver or the kidney is still unsolved for machines.