This is huge, considering how simple the circuitry is.

Based on the prototyping process, they have gone for an expensive impedance analysis DSP chip with 0.1Hz resolution and a relatively wide frequency range. If this gets adopted by a major manufacturer, you could spin an ASIC for pocket change at the optimal frequency (40KHz empirically derived by researchers). Additionally, put the multiplexer on the same chip and make a single chip solution.

The paper is focused more on the machine learning aspect but there are obvious points to improve on the electrical bits!

I'll be waiting until this appears as a Myo-like form factors for hackers to play with.

This is amazingly cool in general. The idea of a lightweight, wearable sensor that can detect gestures is very exciting. I wonder how hard / practical it will be for somebody else to implement this? I'm almost intrigued enough to take a stab at it.

I don't see how is this an alternative to CT or MRI. It is more likely analogous with regard to the principle of imaging, not an alternative.

It might not be a "competitive" alternative, but it does what CT and MRI do — takes picture inside the body.