Our job is to reduce false positives.
We have two sets of data: Negatives and True Positives.
In theory, I understand, these should be sufficient for the job.
However, there is a subset of the Negative screens that were false positives. If we can access these, I believe they may help with the overall task because, apart simply from reliability of manifestation, there might be common artefacts (or absence of artefacts) that contributed to their false identification as positives. If there are, they could be valuable additions to the training.
Looking at the data dictionary, there is a field 'daysSincePreviousExam'. Normally, screenings would take place annually.
Would it be correct to use this field to identify false positives?
If daysSincePreviousExam < 150, say, as a cut-off well below the expected >360, would we be correct in concluding that this is the scan of a recalled patient?
If so, is there any consistent way to get the previous scan (the one that produced the false positive) - or is that just the 'luck of the draw' if that patient happens to have multiple scans on your system? .... or, is it the case that one common reason for there being more than one set of scans for a patient on your system, is precisely that they are a recalled patient?
Created by Peter Brooks fustbariclation Thanks for your questions!
We decided to only provide information about the objective patient information which are the mammography images, and whether the woman developed (Positive Set) or didn't develop (Negative Set) cancer within a year from the last mammography exam. Information such as breast density, whether the woman was recalled or not, or BIRADS assessment is not given because it is considered a radiologist subjective interpretations of the images.
It wouldn't be correct to interpret daysSincePreviousExam < 150 as a scan of a False Positive. A woman may come in for pain in between negative screening exams, and be deemed to have a benign finding and sent back to routine screening.
Also, it wouldn't be correct to interpret multiple scans as an indication of a recall. Larger woman, e.g., may get more than the usual 2 scans per breast.
Let us know if you need more clarifications.
Gustavo