JordanMicahBennett
commented
4 years ago Hello, is your dataset composed of just those images in data folder? If yes I think that is a dataset too little to train a reliable generalized model, isn't it?
Apart from this I would like to point you to my project, which shares the objectives with yours and has a larger dataset of both RX and CT scans.
You can find it here: https://github.com/AleGiovanardi/covidhelper
I also would like to suggest you to try out the analysis of CT scans instead of RX for three main reasons:
- CT are widely more used for diagnostic purposes.
- Pathological regions are more clear to see and shows clear signs of identification of the disease like you can see here: https://www.auntminnie.com/index.aspx?sec=sup&sub=cto&pag=dis&ItemID=128300
- In CT scans there is less noise compared to RX scans
I want also to congratulate with you for your job, it is an amazing work imho. Keep it up.
Kind regards,
Ale.
Good day, Ale. You'll find that lack of data is a general issue. In my repository, I mentioned this.
Also, about 6 weeks before your inquiry/comment, a few CT scan images were used initially, and this is why the repo had CT scan in the name since Feb 9, 2020, although CT scan image data is scarce.
- I made a request for more data via Chinese researchers, though no response yet.
- I made a call to my Government on February 17, to try to use their administrative powers to liaise with countries with large amounts of Covid lung scan data, to try to acquire said data.
I think the Covid19 dataset I use is too large to be uploaded to github.
-
I had mentioned in the repository in the code setup section, that the dataset (containing 188 images) was available on a google drive link of mine.
-
The data was taken from some Covid19 papers, as well as from a repository by Dr Jospeh Cohen, which contains 200+ images.
-
The goal is to use any type of image data available, in a consistent way. So far, I've been using coronal-view (frontal) based images.
What types of views have you been using?
What is your Sensitivity/Accuracy/Specificity so far?
Take care and regards, and your mother is a brave lady!
Hello, is your dataset composed of just those images in data folder? If yes I think that is a dataset too little to train a reliable generalized model, isn't it?
Apart from this I would like to point you to my project, which shares the objectives with yours and has a larger dataset of both RX and CT scans.
You can find it here: https://github.com/AleGiovanardi/covidhelper
I also would like to suggest you to try out the analysis of CT scans instead of RX for three main reasons: 1) CT are widely more used for diagnostic purposes. 2) Pathological regions are more clear to see and shows clear signs of identification of the disease like you can see here: https://www.auntminnie.com/index.aspx?sec=sup&sub=cto&pag=dis&ItemID=128300 3) In CT scans there is less noise compared to RX scans
I want also to congratulate with you for your job, it is an amazing work imho. Keep it up.
Kind regards,
Ale.