Narrow from field to question
Move from field, to sub-field, to method, to question. 'Machine learning' becomes 'medical imaging', becomes 'segmentation under limited labels', becomes a question you could answer in three or four years.
Test the topic against three checks
Is someone publishing on it now? Is there funding attached to it? Could you describe the first experiment or study you would run? If any answer is no, the topic is not ready.
Let supervisors shape it
Most doctoral topics are negotiated, not chosen alone. Arrive with a clear direction and stay open to the version the group can actually support and fund.
Put this into practice
Search supervisors by country, university and research area, or let the PhD Finder rank funded positions against your profile.