Let's dream of an harmonized world! By the way, the FDA has done a huge work to classify the medical devices. Just have a look at their product classification database and you will see that there are almost no shadow zones, compared to the GHTF rules.
The main difference is that FDA has only 3 levels of classification, with very different paths to certification (premarket notification and premarket approval), compared to CE mark for example.

Classification of software

The document contains a definition of active medical devices, which includes software. And there is a paragraph in structure of the classification rules about how dealing with software: standalone, not standalone, software driving a device and so on. To sum-up:

  1. software are active devices,
  2. if they are not standalone, they fall into the category of the device they influence/drive/monitor/control,
  3. if they are standalone, apply the rules 9 to 12,
  4. the last rule (rule 12) is a fall-trough rule, which gives class A to every device, which doesn't match other rules.

No very different from CE mark or Canadian regulation.

Which classification for Medical Imaging Software?

Now let's have a closer look at Medical Imaging Software or PACS (Picture Archiving Communication System). The GHTF gives samples for every rule. For rule 12, you find: "powered equipment for the recording, processing, viewing of diagnostic images". So every PACS and standalone Medical Imaging Software are class A in the GHTF recommendation.
Not only the viewers are class A but also software processing diagnostic images. Canada and CE didn't have those examples in their guidances. More, they issued notices about medical imaging software giving indications seemingly in contradiction with those of GHTF:

I agree with CE and Canada guidances. Software post-processing images for diagnosis shall be at level II, not level I. So I think that the samples given by the GHTF should be redefined to match the CE and Canadian rules. The GHTF rule 10(i) is about active devices for direct diagnosis. Inserting in the samples the post-processing imaging software to allow diagnosis would be a solution. This is not a big issue but it would clarify the situations of post processing tools.
The harmonzation is ongoing but not for tomorrow!