search

DIAGNijmegen / bodyct-dsb2017-grt123

Repository which contains the code of the grt123 solution from the Kaggle DSB 2017 challenge on lung cancer detection
Other
2 stars 0 forks source link

Cases fail to process with AttributeError #46

Closed cjacobs1 closed 3 years ago

cjacobs1 commented 3 years ago

I have three example cases that fail using the current version of the processor GC (v1.1.2) with the following error: AttributeError: 'NoneType' object has no attribute 'split'

I think this may be because no nodules are detected at all?

I'll put the three cases here: \\umcsanfsclp01.umcn.nl\radng_bodyct\software\LungCAD\grt123-failures

cjacobs1 commented 3 years ago

@Whardian : Could you have a look at this? Would be a good first check at the codebase for you as well.

cjacobs1 commented 3 years ago

CC to @silvandeleemput

cjacobs1 commented 3 years ago

Full log:

2021-02-23T15:44:20.636018299Z THCudaCheck FAIL file=/pytorch/aten/src/THC/THCGeneral.cpp line=844 error=11 : invalid argument
2021-02-23T15:44:33.386017517Z Traceback (most recent call last):
2021-02-23T15:44:33.386051856Z   File "apply.py", line 100, in <module>
2021-02-23T15:44:33.386063002Z     output_json.append(extract_json_report(xmlfile=output_xml))
2021-02-23T15:44:33.386071212Z   File "apply.py", line 20, in extract_json_report
2021-02-23T15:44:33.386079629Z     xmlreport = LungCadReport.from_xml(ElementTree.parse(str(xmlfile)))
2021-02-23T15:44:33.386087955Z   File "/home/user/xmlreport.py", line 110, in from_xml
2021-02-23T15:44:33.386096229Z     cancerinfo = CancerInfo.from_xml(cancerinfo)
2021-02-23T15:44:33.386123579Z   File "/home/user/xmlreport.py", line 367, in from_xml
2021-02-23T15:44:33.386135138Z     referencenoduleids = tuple([int(e) for e in xml.find("./ReferenceNoduleIDs").text.split(',')])
2021-02-23T15:44:33.386143344Z AttributeError: 'NoneType' object has no attribute 'split'
silvandeleemput commented 3 years ago

I think I'll take this issue, as the xmlreport.py should be installed from: https://github.com/DIAGNijmegen/bodyct-nodule-evaluation to avoid code duplication

silvandeleemput commented 3 years ago

@cjacobs1 It is indeed a case of reading an XML report with a CancerInfo section but without any ReferenceNoduleIDs for CancerInfo. Should this be allowed? Does no reference nodules imply a low chance of cancer? Could you provide the failing cases?

cjacobs1 commented 3 years ago

The failing cases are now in the folder I mentioned above. Yes, it is allowed that there are no reference nodule ids. That indeed implies a low chance of cancer, (although not zero because grt123 has a 'dummy' nodule as part of the solution).