On asking thirty-seven times

The methods section of a thesis makes data acquisition sound like a lookup. In practice it was eleven months of asking.

I went to hospitals in person, because email from a student is easy to ignore and a person in a corridor is slightly harder. Every hospital request was declined, which I want to record plainly rather than fold into a passive sentence. The reasons were reasonable — patient consent, no de-identification pipeline, nobody whose job it was to say yes.

I emailed the corresponding author of every paper whose dataset I could use, and went through more than thirty-seven published papers to find them. Most did not reply. The FGADR authors did.

One reply is a dataset. That is Axiom V, and it is not an inspirational line — it is a description of the actual yield rate. If you are a student reading this and your first twenty emails went unanswered, that is the normal case, not evidence that you are doing it wrong.

The part nobody warns you about is that the reply creates work. A licence to use data is the beginning of the preprocessing, not the end of the acquisition.