As you'll know, there are some errors in the test data you've sent us. In particular there are some mis-labelled images: Two images have the entry 'CCID' instead of 'CC', and one has 'CV' instead of 'CC' ( the files are: 578699, 267643 and 267631). My question is whether you know if there are any similar errors in the training data files.

Created by Peter Brooks fustbariclation
Thank you for that - I remember it now. I'd been thinking of the presentation, where I'd got the wrong impression. Also, it seems, almost all views are CC or MLO. Anyway, I'm pleased to be clear about it now, and prepared for unusual views.
Hi Peter, > Though the advice was that the views we had been given would be either CC or MLO. Please refer to the [Data Description](https://www.synapse.org/#!Synapse:syn4224222/wiki/401743) page for additional information on the view types. The list of all the views is available on the third sheet of the [Challenge Dictionary](https://www.synapse.org/#!Synapse:syn7214004).
Thank you, that's very useful - and makes sense. Though the advice was that the views we had been given would be either CC or MLO. Do you know if there are similarly anomalous views in the training data? It's just sensible to know what is there, so, if there are some anomalous views, they can be excluded. In real life, I'd imagine that the radiologist and referring physician would expect to inspect such anomalous images manually.
The images you point out are not labeled incorrectly. Note that file 578699 is correctly labeled "CV" = cleavage view. It is not a CC. Also, files 267643 and 267631 appear to be labeled correctly as "CCID" = craniocaudal implant displaced. You can see the implant (white) posteriorly on one of the views and it is likely pushed/displaced posteriorly enough on the other that it doesn't show up at all on the image.

Data integrity - errors in image files page is loading…