Datasets:
Query Regarding Potential Data Duplication and Naming Convention in CT Dataset
Thank you very much for sharing this interesting and valuable work. We appreciate your significant contribution to the community.While downloading and initially processing the CT nii data, I encountered a potential data consistency issue that I hope you could clarify.
Problem Description:I observed what appears to be data duplication within the dataset. Specifically, I suspect that when the two numbers preceding the first underscore (_) in the filename are identical, the corresponding data files contain the exact same CT image.For example:3_5_155.nii.gz 3_5_41.nii.gz3_5_194.nii.gz. My preliminary check indicates that the CT image data content in these three files (and potentially other similarly named files）appears to be completely identical.
If my observation is correct, this raises questions about the data organization and naming convention:
If multiple filenames correspond to the exact same CT image, how are the associated eye-tracking points （or any other separately annotated data）correctly matched to a specific file? When I attempted to match the eye-tracking data file by file, I found that the information did not seem to align with these seemingly duplicated CT images.
Thank you for your time and assistance!
This is my visualization result ： 3_5_194 There is an endotracheal tube with a tip just below the thoracic inlet.
Hi, thank you for bringing it up. Do you visit iccv right now? We can discuss.
It's a huge bummer I couldn't make it to ICCV. I really wish I could talk to the authors in person.
Yesterday, I re-sorted the data and articles. I've observed that files sharing the same prefix (e.g., $\text{3_5}$ in $\text{3_5_155}$ and $\text{3_5_41}$) contain the exact same CT image data. The different suffixes (e.g., $\text{155}$ and $\text{41}$) likely refer to reconstruction parameters (e.g., slice thickness) for the same patient case.
However, I am still unable to accurately match the Eye-tracking Data and the corresponding Text Annotations to the correct CT} image files. Any guidance on the data processing workflow, especially on how to achieve a precise link between eye-tracking/text and the images, would be greatly helpful.
Hi. That's a bummer because I really want to discuss with an actual user to get more viewpoints.
Yes, we have many shared CTs for the same patient case, hence the same report. This is because we want to strictly follow what radiologists actually do in practice: the technician captures a full detail CT 1mm and then processes it into multiple thickness variations and passes them to the experts. We recorded multiple gaze sequence for the same CT with different thicknesses so that the number of scanpath-CT pairs in the dataset is richer. The text transcripts are not fine-grained right now because they were directly extracted from Google Speech-to-Text. I suspect there may be mismatches between the timestamps and the text. Would you like the audio data? I can share it via email if you leave your email in your reply. I believe that would help greatly if you want to directly align text (or even audio) to gaze points.
That is incredibly helpful, thank you! Since I'm also conducting eye-tracking studies, I have many similar questions and would greatly appreciate the chance to connect and communicate with someone working in the same direction.
Please feel free to reach me at: [email protected]
Let's discuss this further!