-- the superconducting magnet shell requires the accurate creation of 'nested doll' cryogenic containers, built to withstand magnetostatic forces in a highly regulated environment with defined safety requirements under catastrophic failure modes. Solving the design problem is equivalent to solving a huge set of nasty, coupled PDEs subject to loads of material constraints. This is directly analogous to aspects of jet engine design.
-- inside the bore of the magnet (but not in the cryostat) goes a device called the gradient set, whose job is to generate \partial B_z/ \partial_{xyz} as a function of time (that, very much indirectly, the radiographer specifies). This is a water cooled, resistive set of magnet coils with a defined frequency response curve, linearity requirements, etc. The current into them is generated by a set of three huge amplifiers, which have to actually take a signal delivered on a timebase of microseconds and volts and amplify it with negligible delay and deliver kA into a large inductor centimetres away from a patient. This is a formidable (power) electronic engineering challenge with huge parallels to various aspects of electrical engineering – e.g. managing (preventing) dielectric breakdown, thermal management, inverse solutions to Maxwell's equations in a quasistatic region (people use streamfunctions to do this well), etc.
-- the RF side of the system has to transmit kV and receive microvolts within microseconds into a definitively challenging electrodynamic environment with constraints on harmonics. Everything has to keep to a hard realtime constraint. The ADC must have a huge dynamic range and the problem is conducted massively in parallel. This is directly analogous to problems in telecommunications or RF design, but harder -- intermittent pulsed not continuous wave, and a hard requirement to accurately measure analogue voltages. Designing the RF coil ("probe" in NMR speak or ≈"antenna") is a further horrible (full-wave) EM design problem that even GE often subcontract out to one of about five specialist firms worldwide.
It's not a priori obvious to me that lots of competing companies would be better at creating stuff that requires the interaction of disciplines like this. Rather, I view the split up of GE and all of the woes of the article as evidence of a business mismanaged by MBAs. The defined benefit pensions should have been protected by law and overseen by an independent regulator - like my defined benefit pension that sits above my employer and shares risk among many different universities.
I guess what I'm trying to say is: it's almost as if the combination of supporting physics and engineering without particular attention to the ultimate "core competency" or "end market segment" lets creative, interdisciplinary ideas like this flourish.