A lot depends on what you mean by "duplicates". When you import and the option to "Don't Import Suspected Duplicates" is checked, LrC will check the filename, filesize and capture date of the potential imports and compare them to images already in the catalog. In which case, a "duplicate" has to be an identical copy of a cataloged file, and it it is not an identical copy then LrC will try to import it. So for example, a jpeg might have been created by exporting a raw file, and although the image is the same the file characteristics are not....therefore it's not a duplicate as far as LrC is concerned. There can be many examples of files containing the same visual image which would not be regarded as "duplicates", which may well explain the problem with those 50k images.
You appear to be sure that there are only 5k out of those 69k files which are not duplicates, so the obvious question is how do you know that? Following on from that, are you able to find and isolate those 5k images outside LrC, so that you can specifically target only those 5k files for import?