search

uci-cbcl / DeepLung

WACV18 paper "DeepLung: Deep 3D Dual Path Nets for Automated Pulmonary Nodule Detection and Classification"
GNU General Public License v3.0
153 stars 54 forks source link

about 'annotationdetclsconvfnl_v3.csv' file in your repository #6

Closed HunarAA closed 6 years ago

HunarAA commented 6 years ago

Hi, I am working on a project to classify lung CT scan (cancer/no-cancer) on luna16 dataset using CNN. Is your file 'annotationdetclsconvfnl_v3.csv' can be used for that purpose? because the file contains cancer class. where did you get this file? because both LIDC-IDRI and luna16 dataset does not have this file. have you created it by yourself based on nodule diameter? can you tell me more detail

wentaozhu commented 6 years ago

Yes, you can. It is generated by DeepLung/nodcls/data/dataconverter.py

LUNA16 comes from LIDC-IDRI. LUNA16 has nodule detection ground truth. LIDC-IDRI has nodule classification label. We find the correspondence using extclsshpinfo.py

Also refer #4 for the finding correspondence.

HunarAA commented 6 years ago

I am very thankfull for your answer, I want to use it in my master research of i will cite your paper. Can you send the csv file for me? Also can you tell me what is the base that you decide the class (0 or 1)? Its based on nodule size only? Or what else? Wednesday, 12 September 2018, 07:34pm +03:00 from Wentao Zhu notifications@github.com :

Yes, you can. It is generated by DeepLung/nodcls/data/dataconverter.py LUNA16 comes from LIDC-IDRI. LUNA16 has nodule detection ground truth. LIDC-IDRI has nodule classification label. We find the correspondence using extclsshpinfo.py Also refer #4 for the finding correspondence. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub , or mute the thread .

wentaozhu commented 6 years ago

You can read the paper for details. You can download the csv from the github.

HunarAA commented 6 years ago

Ok Thank you, I read the paper but in data section there is not a much detail about how you choose the classes. Kindly can you explain it for me?

Sent from myMail for Android Wednesday, 12 September 2018, 10:53pm +03:00 from Wentao Zhu notifications@github.com :

You can read the paper for details. You can download the csv from the github. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub , or mute the thread .

HunarAA commented 6 years ago

Hi, hope you answer my question: after working on the annotation file 'annotationdetclsconvfnl_v3.csv' I saw that there is only 545 ID's out of 888 scans (as you know each scan has a unique ID) in luna16 dataset, why? where are other 343 scans (ID's)? why do you not include them in your annotation file? Is there a reason for that? please, explain it to me? regards...hunar

On Wed, Sep 12, 2018 at 10:53 PM Wentao Zhu notifications@github.com wrote:

You can read the paper for details. You can download the csv from the github.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/uci-cbcl/DeepLung/issues/6#issuecomment-420775885, or mute the thread https://github.com/notifications/unsubscribe-auth/AkvxrGgbhzlMRuHmqtxYWtfpoiS9LLANks5uaWYpgaJpZM4Wku2o .

-- Hunar A.Ahmed Mobile: +9647705493448 email: honar.cs@gmail.com address: Iraq, Kurdistan.

wentaozhu commented 6 years ago

How many positive CT images in LUNA 16? I remember not all is positive.

HunarAA commented 6 years ago

let me explain it more, as you know all records in ' annotationdetclsconvfnl_v3.csv' are 1004, but only 545 of them are unique ids others are repeated ones, of all 1004, 450 are positive and 554 are negative. as you know luna16 has 888 scans (each scan has a unique ID) but in your annotation file there are only 545 unique IDs, means that there are other 343 unique IDs of total 888 luna16 scans.

On Sun, Sep 16, 2018 at 8:17 PM Wentao Zhu notifications@github.com wrote:

How many positive CT images in LUNA 16? I remember not all is positive.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/uci-cbcl/DeepLung/issues/6#issuecomment-421798747, or mute the thread https://github.com/notifications/unsubscribe-auth/AkvxrGr2Pdkf8PuFUvTg-lllwHY4TYh6ks5uboe5gaJpZM4Wku2o .

-- Hunar A.Ahmed Mobile: +9647705493448 email: honar.cs@gmail.com address: Iraq, Kurdistan.

wentaozhu commented 6 years ago

Yes. Each positve CT image consisting of nodules consists of multiple nodules.

HunarAA commented 6 years ago

dear friend, this is not the answer to my question, I know that each scan may consist of no nodule, one nodule or more than one nodule, I didn't ask you this question. there are scans that not include in your annotation file, only tell me why these scans not included? I saw that some scans are also not included in luna16 annotation.csv file.

On Mon, Sep 17, 2018 at 10:33 AM Wentao Zhu notifications@github.com wrote:

Yes. Each positve CT image consisting of nodules consists of multiple nodules.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/uci-cbcl/DeepLung/issues/6#issuecomment-421914952, or mute the thread https://github.com/notifications/unsubscribe-auth/AkvxrB9iCIm-MjAqaPJWEhjUXTBBqw4Lks5ub1BUgaJpZM4Wku2o .

-- Hunar A.Ahmed Mobile: +9647705493448 email: honar.cs@gmail.com address: Iraq, Kurdistan.

wentaozhu commented 6 years ago

The negative CT image consisting no nodules shouldn't be included in the classification label files.

HunarAA commented 6 years ago

thank you very much, at last, I get the answer, how much your classification accuracy? did you shuffle the data? are you did augmentation for the data?

On Mon, Sep 17, 2018 at 3:11 PM Wentao Zhu notifications@github.com wrote:

The negative CT image consisting no nodules shouldn't be included in the classification label files.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/uci-cbcl/DeepLung/issues/6#issuecomment-421989325, or mute the thread https://github.com/notifications/unsubscribe-auth/AkvxrMGGHoaF6SurCZ-ok1USCJ77jrLvks5ub5GEgaJpZM4Wku2o .

-- Hunar A.Ahmed Mobile: +9647705493448 email: honar.cs@gmail.com address: Iraq, Kurdistan.

wentaozhu commented 6 years ago

Yes. Yes. Congratulations!

HunarAA commented 6 years ago

dear, you don't tell me what is the base that you decide the class (0 or 1)? Its based on nodule size only? Or something else? I use your annotation for my master thesis and I should know the annotation process?

On Mon, Sep 17, 2018 at 6:51 PM Wentao Zhu notifications@github.com wrote:

Yes. Yes. Congratulations!

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/uci-cbcl/DeepLung/issues/6#issuecomment-422068876, or mute the thread https://github.com/notifications/unsubscribe-auth/AkvxrDwNDsb0MY-PcSmFRXOfWzXxbDg1ks5ub8USgaJpZM4Wku2o .

-- Hunar A.Ahmed Mobile: +9647705493448 email: honar.cs@gmail.com address: Iraq, Kurdistan.

wentaozhu commented 6 years ago

Please read the paper for details