{ "id": 13259, "title": "Unrecognizable distinctions between stage 0 and stage 2?", "url": "/competitions/diabetic-retinopathy-detection/discussion/13259", "author_name": "", "post_date": "2015-04-06T23:59:57.603Z", "votes": 1, "comment_count": 24, "views": 5754, "content": "

As I was studying the possible distinction between DR stages and the pictures belong to those, I found that I did not find any distinction between pictures 40_right and 40_left, meanwhile the label file shows stage 0 and stage 2 respectively. Perhaps it is my bad?

", "messages": [ { "id": "69846", "postDate": "04/06/2015 23:59:57", "content": "

As I was studying the possible distinction between DR stages and the pictures belong to those, I found that I did not find any distinction between pictures 40_right and 40_left, meanwhile the label file shows stage 0 and stage 2 respectively. Perhaps it is my bad?

", "rawMarkdown": "", "votes": null }, { "id": "69858", "postDate": "04/07/2015 05:48:50", "content": "

I will have a closer look at the mentioned pictures later as I get home. Till that I can say, that there are some mis-classified images: for example having DR stage 0 although there are some defininte microaneurysms on them.

\n

The data description says: "Like any real-world data set, you will encounter noise in both the images and labels." So it is also possible that sometimes the classification is not correct.

", "rawMarkdown": "", "votes": null }, { "id": "69867", "postDate": "04/07/2015 08:09:47", "content": "

Thank you Zoltan.

\n

Are you a specialist in diabetes?

", "rawMarkdown": "", "votes": null }, { "id": "69868", "postDate": "04/07/2015 09:25:08", "content": "

No, not really. I have just some insight into this area as my MSc degree thesis was about computer aided DR detection on retinal images including a gradient based vessel and microaneurysm detector algorithm. (The microaneurysm detector achieved the 3rd place on the ROC: http://ophth.healthcare.uiowa.edu/ROC/var.1/www/results.php, Z. Fegyver)

\n

That is the reason why I am interested in this competition. To tell you the truth, I don't know how far I can get as I am not experienced in CNNs and similar methods but in "classic" processing of retinal images. We will see. :)

", "rawMarkdown": "", "votes": null }, { "id": "69874", "postDate": "04/07/2015 11:52:21", "content": "

I noticed that 10996_left is all black? How do you handle that?

", "rawMarkdown": "", "votes": null }, { "id": "69875", "postDate": "04/07/2015 12:26:02", "content": "

[quote=Zoltan Fegyver;69868]

\n

No, not really. I have just some insight into this area as my MSc degree thesis was about computer aided DR detection on retinal images including a gradient based vessel and microaneurysm detector algorithm. (The microaneurysm detector achieved the 3rd place on the ROC: http://ophth.healthcare.uiowa.edu/ROC/var.1/www/results.php, Z. Fegyver)

\n

That is the reason why I am interested in this competition. To tell you the truth, I don't know how far I can get as I am not experienced in CNNs and similar methods but in "classic" processing of retinal images. We will see. :)

\n

[/quote]

\n

Congrats(Gratulálok:))) Zoltán, nice result, we will see, yes :)

\n

and your opinion to picture 40?

", "rawMarkdown": "", "votes": null }, { "id": "69876", "postDate": "04/07/2015 12:40:48", "content": "

[quote=skaae;69874]

\n

I noticed that 10996_left is all black? How do you handle that?

\n

[/quote]

\n

Yes, it can be the the case, where the phisician made the serious mistake and looked into the instrument with his closed eye. :)

\n

Seriously, I don't know that whether we should handle somehow this kind of extrem sample.

", "rawMarkdown": "", "votes": null }, { "id": "69877", "postDate": "04/07/2015 12:46:35", "content": "

Thanks. I think i'll just dublicate the other eye.

", "rawMarkdown": "", "votes": null }, { "id": "69879", "postDate": "04/07/2015 12:57:33", "content": "

[quote=skaae;69877]

\n

Thanks. I think i'll just dublicate the other eye.

\n

[/quote]

\n

Some filtering may help us to obtain a tiny information from this picture

", "rawMarkdown": "", "votes": null }, { "id": "69902", "postDate": "04/07/2015 16:45:25", "content": "

@George Solymosi

\n

Thanks! :)

\n

I had a look at the #40 images, I would classify both of them as stage 1 as they contain several microaneurysms, some of them are marked by red arrows. On the 40_left image I marked two hemorrhage-like objects by white arrows, but I am not sure because they are too bright. If they are really hemorrhages, the stage of that image shall be 2 instead of 1. (The images are the outputs of my current image preprocessor.)

\n

@Skaae

\n

I have attached my preprocessed variant of image 10996_left. It seems almost unusable.

", "rawMarkdown": "", "votes": null }, { "id": "69909", "postDate": "04/07/2015 17:47:44", "content": "

Ok. I didnt even notice that there was anything on it.

\n

How do you process the images? or is that confidential :)

", "rawMarkdown": "", "votes": null }, { "id": "69983", "postDate": "04/08/2015 08:10:04", "content": "

[quote=skaae;69909]

\n

Ok. I didnt even notice that there was anything on it.

\n

How do you process the images? or is that confidential :)

\n

[/quote]

\n

The preprocessing is still under construction, so I would keep it as a secret till some time. :)

", "rawMarkdown": "", "votes": null }, { "id": "70017", "postDate": "04/08/2015 14:43:47", "content": "

Hi Zoltan, thanks for sharing your insights.

\n

It seems like the objects you marked with white arrows are possibly dirt as well, since they look to be visible in 40_right as well. This is an interesting complication which I hadn't considered before.

\n

In 10966, those small rounds blotches covering most of the image seem like a pattern apparent on some of the stage 4 images, would you agree? Do you know the name of that phenomenon?

", "rawMarkdown": "", "votes": null }, { "id": "70029", "postDate": "04/08/2015 17:11:39", "content": "

Dear Emolson, thanks for your comment, I had not recognized those objects on the 40_right image: they are really dirt... I noticed these kind of dirty spots on several images, I think, this really should be taken into consideration as sometimes they have very similar appearance compared to microaneurysms and hemorrhages.

\n

I am not sure about those smaller, bright objects you mentioned as their shape are too regular, they are almost ellipses. Bright-colored indicators of DR are: cotton-wool spots and hard/soft exudates, the presence of them indicates a higher stage of DR. But usually they have much more irregular shape, just search on Google for sample images. I will have a closer look at these objects you mentioned as I will have enough time. But because of their shape I would say for now that they are only "bright-dirt" on some images due to some reason. :)

", "rawMarkdown": "", "votes": null }, { "id": "70057", "postDate": "04/08/2015 22:00:45", "content": "

Have a look at images 10640_left and 10643_left. On both images there are the same kind of bright spots, but both of them are marked as stage 0. It seems, these kind of spots are "dirt" too.

", "rawMarkdown": "", "votes": null }, { "id": "70082", "postDate": "04/09/2015 04:29:37", "content": "

Hello Zoltan,
Is it ok if I present you some more images in this thread ?
I have a number of cases too that I am confused about.

I tried to find some info on the internet but it's very hard to find some definitive answers..

", "rawMarkdown": "", "votes": null }, { "id": "70099", "postDate": "04/09/2015 07:42:02", "content": "

Hi Julian, of course feel free to share more images, I think it would be useful to discuss some confusing cases as there are many of them... Congrats on your score on the leaderboard!

", "rawMarkdown": "", "votes": null }, { "id": "70140", "postDate": "04/09/2015 18:53:21", "content": "

Of course I cannot find the examples right now.. :S
However.. sometimes an eye gets rated a '2' and I barely see any irregularities.

Could it be that there is also something to see in the vein structuring (apart from wildgrowth) ?

I read about meandering but see no structure in the ratings when I see meandering veins.
Also I read about unnatural angles after 2 cross each other. But I've never seen/reckognized that.


", "rawMarkdown": "", "votes": null }, { "id": "70141", "postDate": "04/09/2015 19:02:19", "content": "

Here is a 2 where I cannot see anything.

", "rawMarkdown": "", "votes": null }, { "id": "70149", "postDate": "04/09/2015 20:17:11", "content": "

I have found a microaneurysm on it, marked by red. What is suspicious for me is the possibility of neovascularization. I have marked some places where it may be seen but as I am not a specialist, I may fail this point. Tomorrow I will read some publications about changes in the vessels due to DR.

", "rawMarkdown": "", "votes": null }, { "id": "70151", "postDate": "04/09/2015 20:23:05", "content": "

You are right, the meander effect also counts, but this parameter seems to be rather subjective -- at least for me.

\n

Also I have not found a direct relation between DR stages and the meander effect. Tortuosity is another word for describing this effect, possibly you may find some more useful publications using this term.

", "rawMarkdown": "", "votes": null }, { "id": "71445", "postDate": "04/14/2015 04:36:02", "content": "

@Zoltan

\n

Thank you by the way.

\n

I do have many more examples but when I have time I'm planning to post them somewhere.


", "rawMarkdown": "", "votes": null }, { "id": "71471", "postDate": "04/14/2015 09:58:35", "content": "

Dear Julian, I still had no time for having a deeper look at more publications in this topic... I hope, this will change in the following days. Feel free to post your examples here!

", "rawMarkdown": "", "votes": null }, { "id": "73040", "postDate": "04/21/2015 22:57:47", "content": "

Hmm.. The 10397_left.jpg  has level 0 and the 10397_right.jpg  has level 2. But I cannot recognize distinction except some blurring, because.... it's the same  right  eye (level 2)

\n\n

The same is with 10363_xxx.jpg case.

\n

Is it a property of the  "real world" data or I just have an issue with my local uploaded data?

", "rawMarkdown": "", "votes": null }, { "id": "77766", "postDate": "05/10/2015 14:18:56", "content": "

[quote=Roman Gudchenko;73040]

\n

Hmm.. The 10397_left.jpg  has level 0 and the 10397_right.jpg  has level 2. But I cannot recognize distinction except some blurring, because.... it's the same  right  eye (level 2)

\n

The same is with 10363_xxx.jpg case.

\n

Is it a property of the  "real world" data or I just have an issue with my local uploaded data?

\n

[/quote]

\n

That's amazing. The 10397 left/right are obviously identical eyes--look at the vasculature. Yet one is rated 0 and the other 2! Makes me wonder about the reliability of the "clinical rater" for the study.

\n

Can anyone explain the difference in rating?

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 69858, "author_name": "zoltanfegyver", "author_url": "", "post_date": "04/07/2015 05:48:50", "content": "

I will have a closer look at the mentioned pictures later as I get home. Till that I can say, that there are some mis-classified images: for example having DR stage 0 although there are some defininte microaneurysms on them.

\n

The data description says: "Like any real-world data set, you will encounter noise in both the images and labels." So it is also possible that sometimes the classification is not correct.

", "votes": null, "replies": [] }, { "id": 69867, "author_name": "", "author_url": "", "post_date": "04/07/2015 08:09:47", "content": "

Thank you Zoltan.

\n

Are you a specialist in diabetes?

", "votes": null, "replies": [] }, { "id": 69868, "author_name": "zoltanfegyver", "author_url": "", "post_date": "04/07/2015 09:25:08", "content": "

No, not really. I have just some insight into this area as my MSc degree thesis was about computer aided DR detection on retinal images including a gradient based vessel and microaneurysm detector algorithm. (The microaneurysm detector achieved the 3rd place on the ROC: http://ophth.healthcare.uiowa.edu/ROC/var.1/www/results.php, Z. Fegyver)

\n

That is the reason why I am interested in this competition. To tell you the truth, I don't know how far I can get as I am not experienced in CNNs and similar methods but in "classic" processing of retinal images. We will see. :)

", "votes": null, "replies": [] }, { "id": 69874, "author_name": "skaaes", "author_url": "", "post_date": "04/07/2015 11:52:21", "content": "

I noticed that 10996_left is all black? How do you handle that?

", "votes": null, "replies": [] }, { "id": 69875, "author_name": "", "author_url": "", "post_date": "04/07/2015 12:26:02", "content": "

[quote=Zoltan Fegyver;69868]

\n

No, not really. I have just some insight into this area as my MSc degree thesis was about computer aided DR detection on retinal images including a gradient based vessel and microaneurysm detector algorithm. (The microaneurysm detector achieved the 3rd place on the ROC: http://ophth.healthcare.uiowa.edu/ROC/var.1/www/results.php, Z. Fegyver)

\n

That is the reason why I am interested in this competition. To tell you the truth, I don't know how far I can get as I am not experienced in CNNs and similar methods but in "classic" processing of retinal images. We will see. :)

\n

[/quote]

\n

Congrats(Gratulálok:))) Zoltán, nice result, we will see, yes :)

\n

and your opinion to picture 40?

", "votes": null, "replies": [] }, { "id": 69876, "author_name": "", "author_url": "", "post_date": "04/07/2015 12:40:48", "content": "

[quote=skaae;69874]

\n

I noticed that 10996_left is all black? How do you handle that?

\n

[/quote]

\n

Yes, it can be the the case, where the phisician made the serious mistake and looked into the instrument with his closed eye. :)

\n

Seriously, I don't know that whether we should handle somehow this kind of extrem sample.

", "votes": null, "replies": [] }, { "id": 69877, "author_name": "skaaes", "author_url": "", "post_date": "04/07/2015 12:46:35", "content": "

Thanks. I think i'll just dublicate the other eye.

", "votes": null, "replies": [] }, { "id": 69879, "author_name": "", "author_url": "", "post_date": "04/07/2015 12:57:33", "content": "

[quote=skaae;69877]

\n

Thanks. I think i'll just dublicate the other eye.

\n

[/quote]

\n

Some filtering may help us to obtain a tiny information from this picture

", "votes": null, "replies": [] }, { "id": 69902, "author_name": "zoltanfegyver", "author_url": "", "post_date": "04/07/2015 16:45:25", "content": "

@George Solymosi

\n

Thanks! :)

\n

I had a look at the #40 images, I would classify both of them as stage 1 as they contain several microaneurysms, some of them are marked by red arrows. On the 40_left image I marked two hemorrhage-like objects by white arrows, but I am not sure because they are too bright. If they are really hemorrhages, the stage of that image shall be 2 instead of 1. (The images are the outputs of my current image preprocessor.)

\n

@Skaae

\n

I have attached my preprocessed variant of image 10996_left. It seems almost unusable.

", "votes": null, "replies": [] }, { "id": 69909, "author_name": "skaaes", "author_url": "", "post_date": "04/07/2015 17:47:44", "content": "

Ok. I didnt even notice that there was anything on it.

\n

How do you process the images? or is that confidential :)

", "votes": null, "replies": [] }, { "id": 69983, "author_name": "zoltanfegyver", "author_url": "", "post_date": "04/08/2015 08:10:04", "content": "

[quote=skaae;69909]

\n

Ok. I didnt even notice that there was anything on it.

\n

How do you process the images? or is that confidential :)

\n

[/quote]

\n

The preprocessing is still under construction, so I would keep it as a secret till some time. :)

", "votes": null, "replies": [] }, { "id": 70017, "author_name": "emolson", "author_url": "", "post_date": "04/08/2015 14:43:47", "content": "

Hi Zoltan, thanks for sharing your insights.

\n

It seems like the objects you marked with white arrows are possibly dirt as well, since they look to be visible in 40_right as well. This is an interesting complication which I hadn't considered before.

\n

In 10966, those small rounds blotches covering most of the image seem like a pattern apparent on some of the stage 4 images, would you agree? Do you know the name of that phenomenon?

", "votes": null, "replies": [] }, { "id": 70029, "author_name": "zoltanfegyver", "author_url": "", "post_date": "04/08/2015 17:11:39", "content": "

Dear Emolson, thanks for your comment, I had not recognized those objects on the 40_right image: they are really dirt... I noticed these kind of dirty spots on several images, I think, this really should be taken into consideration as sometimes they have very similar appearance compared to microaneurysms and hemorrhages.

\n

I am not sure about those smaller, bright objects you mentioned as their shape are too regular, they are almost ellipses. Bright-colored indicators of DR are: cotton-wool spots and hard/soft exudates, the presence of them indicates a higher stage of DR. But usually they have much more irregular shape, just search on Google for sample images. I will have a closer look at these objects you mentioned as I will have enough time. But because of their shape I would say for now that they are only "bright-dirt" on some images due to some reason. :)

", "votes": null, "replies": [] }, { "id": 70057, "author_name": "zoltanfegyver", "author_url": "", "post_date": "04/08/2015 22:00:45", "content": "

Have a look at images 10640_left and 10643_left. On both images there are the same kind of bright spots, but both of them are marked as stage 0. It seems, these kind of spots are "dirt" too.

", "votes": null, "replies": [] }, { "id": 70082, "author_name": "juliandewit", "author_url": "", "post_date": "04/09/2015 04:29:37", "content": "

Hello Zoltan,
Is it ok if I present you some more images in this thread ?
I have a number of cases too that I am confused about.

I tried to find some info on the internet but it's very hard to find some definitive answers..

", "votes": null, "replies": [] }, { "id": 70099, "author_name": "zoltanfegyver", "author_url": "", "post_date": "04/09/2015 07:42:02", "content": "

Hi Julian, of course feel free to share more images, I think it would be useful to discuss some confusing cases as there are many of them... Congrats on your score on the leaderboard!

", "votes": null, "replies": [] }, { "id": 70140, "author_name": "juliandewit", "author_url": "", "post_date": "04/09/2015 18:53:21", "content": "

Of course I cannot find the examples right now.. :S
However.. sometimes an eye gets rated a '2' and I barely see any irregularities.

Could it be that there is also something to see in the vein structuring (apart from wildgrowth) ?

I read about meandering but see no structure in the ratings when I see meandering veins.
Also I read about unnatural angles after 2 cross each other. But I've never seen/reckognized that.


", "votes": null, "replies": [] }, { "id": 70141, "author_name": "juliandewit", "author_url": "", "post_date": "04/09/2015 19:02:19", "content": "

Here is a 2 where I cannot see anything.

", "votes": null, "replies": [] }, { "id": 70149, "author_name": "zoltanfegyver", "author_url": "", "post_date": "04/09/2015 20:17:11", "content": "

I have found a microaneurysm on it, marked by red. What is suspicious for me is the possibility of neovascularization. I have marked some places where it may be seen but as I am not a specialist, I may fail this point. Tomorrow I will read some publications about changes in the vessels due to DR.

", "votes": null, "replies": [] }, { "id": 70151, "author_name": "zoltanfegyver", "author_url": "", "post_date": "04/09/2015 20:23:05", "content": "

You are right, the meander effect also counts, but this parameter seems to be rather subjective -- at least for me.

\n

Also I have not found a direct relation between DR stages and the meander effect. Tortuosity is another word for describing this effect, possibly you may find some more useful publications using this term.

", "votes": null, "replies": [] }, { "id": 71445, "author_name": "juliandewit", "author_url": "", "post_date": "04/14/2015 04:36:02", "content": "

@Zoltan

\n

Thank you by the way.

\n

I do have many more examples but when I have time I'm planning to post them somewhere.


", "votes": null, "replies": [] }, { "id": 71471, "author_name": "zoltanfegyver", "author_url": "", "post_date": "04/14/2015 09:58:35", "content": "

Dear Julian, I still had no time for having a deeper look at more publications in this topic... I hope, this will change in the following days. Feel free to post your examples here!

", "votes": null, "replies": [] }, { "id": 73040, "author_name": "goodok", "author_url": "", "post_date": "04/21/2015 22:57:47", "content": "

Hmm.. The 10397_left.jpg  has level 0 and the 10397_right.jpg  has level 2. But I cannot recognize distinction except some blurring, because.... it's the same  right  eye (level 2)

\n\n

The same is with 10363_xxx.jpg case.

\n

Is it a property of the  "real world" data or I just have an issue with my local uploaded data?

", "votes": null, "replies": [] }, { "id": 77766, "author_name": "bobalv", "author_url": "", "post_date": "05/10/2015 14:18:56", "content": "

[quote=Roman Gudchenko;73040]

\n

Hmm.. The 10397_left.jpg  has level 0 and the 10397_right.jpg  has level 2. But I cannot recognize distinction except some blurring, because.... it's the same  right  eye (level 2)

\n

The same is with 10363_xxx.jpg case.

\n

Is it a property of the  "real world" data or I just have an issue with my local uploaded data?

\n

[/quote]

\n

That's amazing. The 10397 left/right are obviously identical eyes--look at the vasculature. Yet one is rated 0 and the other 2! Makes me wonder about the reliability of the "clinical rater" for the study.

\n

Can anyone explain the difference in rating?

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "69846": "", "69858": "", "69867": "", "69868": "", "69874": "", "69875": "", "69876": "", "69877": "", "69879": "", "69902": "", "69909": "", "69983": "", "70017": "", "70029": "", "70057": "", "70082": "", "70099": "", "70140": "", "70141": "", "70149": "", "70151": "", "71445": "", "71471": "", "73040": "", "77766": "" }, "source": "meta" }