Open howff opened 1 year ago
We need a way to detect scanned forms which are not clinical images. Typically they contain a lot of text and, even though this is detected and redacted by OCR, there is a risk that some is handwritten (even signatures) and is not redacted.
During CR,DX analysis some rules were found https://git.ecdf.ed.ac.uk/SMI/service/-/issues/188
During MG analysis we discovered that it's not so easy to distinguish forms because of a lack of consistency. More investigation is required.
A ML model has been trained to detect scanned forms, but it will need to be retrained or finetuned on the new MG data.
https://github.com/SMI/dicompixelanon/blob/main/src/testing/learn_forms.md
We need a way to detect scanned forms which are not clinical images. Typically they contain a lot of text and, even though this is detected and redacted by OCR, there is a risk that some is handwritten (even signatures) and is not redacted.
During CR,DX analysis some rules were found https://git.ecdf.ed.ac.uk/SMI/service/-/issues/188
During MG analysis we discovered that it's not so easy to distinguish forms because of a lack of consistency. More investigation is required.
A ML model has been trained to detect scanned forms, but it will need to be retrained or finetuned on the new MG data.
https://github.com/SMI/dicompixelanon/blob/main/src/testing/learn_forms.md