From f0b3b345bac34218b6f222c698ad49f9eb5e0c02 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste Date: Thu, 21 Mar 2024 11:31:38 +0100 Subject: [PATCH 01/28] initialised the lesion_segmentation_protocol.md file based on comment from issue 77 on canproco repo --- lesion_segmentation_protocol.md | 31 +++++++++++++++++++++++++++++++ 1 file changed, 31 insertions(+) create mode 100644 lesion_segmentation_protocol.md diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md new file mode 100644 index 0000000..2969e34 --- /dev/null +++ b/lesion_segmentation_protocol.md @@ -0,0 +1,31 @@ +# Lesion segmentation protocol: + +The following details the protocol for Multiple Sclerosis (MS) lesion segmentation in the spinal cord. +Imaging the spatial cord is often essential to confirm the diagnosis of MS. That is because the lesions of the spinal cord are included in the McDonald diagnosis criteria, which studies dissemination in space and in time [(Thompson et al. 2018)](https://pubmed.ncbi.nlm.nih.gov/29275977/). While the MAGNIMS-CMSC-NAIMS working group recommends to use at least two sagittal images for MS diagnosis, still, axial imaging is mentioned as optional in international imaging guidelines [(Wattjes et al. 2021)](https://pubmed.ncbi.nlm.nih.gov/34139157/). +For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and STIR contrasts. New studies [(Peters et al. 2024)](https://pubmed.ncbi.nlm.nih.gov/38289376/)[(Fechner et al. 2019)](https://pubmed.ncbi.nlm.nih.gov/30679225/) showed that using PSIR contrasts improved MS lesion detection in the spinal cord. [(Fechner et al. 2019)](https://pubmed.ncbi.nlm.nih.gov/30679225/) showed that the PSIR contrast showed a higher signal-to-noise (SNR) ratio compared to the STIR contrast. + +## Criteria to segment MS lesions in the spinal cord: + +Do not segment lesions in images with too many artifacts (such as this one : https://github.com/ivadomed/canproco/issues/53#issue-1938136790). Preferably, add the image to the exclude file so that it isn’t used for model training… +When segmenting lesions on thick slices, always look at the above/below slices to build the volume of the lesion (this can minimize partial volume effect). +Do not segment lesions above the first vertebrae (because here we focus only on MS lesions in the spinal cord). +For lesions segmentations which you are not 100% sure, flag the subject and report it for external validation of the segmentation. + +## How to manually segment lesions: + +- MS spinal cord lesions can be manually corrected from the prediction of a segmentation model or manually segmented from scratch. In the first case, make sure to build the json file associated with the segmentation prediction such as detailed in this comment (https://github.com/ivadomed/canproco/issues/73#issuecomment-1912302223). +- For manual correction of the segmentation file use the manual-correction (https://github.com/spinalcordtoolbox/manual-correction) repository. The command can be inspired from this: + +```console +python manual_correction.py -path-img ~/data/canproco -config ~/config_seg.yml -path-label ~/data/canproco/derivatives/labels -suffix-files-lesion _lesion-manual -fsleyes-dr="-40,70" +``` + +- Then a QC should be produced (prefarably using SCT) and added to a Github issue for further validation by other investigators. + +- If you are not sure of a subject, it should be flagged on Github for a more open discussion. + +## More details: +The following section details the different types of errors which occur during lesion segmentation. It is based on the condensed Nascimento Taxonomy: + +nascimento_taxonomy + From 973c5b25ad32436cb77942b3889158a86e37c00c Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste Date: Thu, 21 Mar 2024 11:35:20 +0100 Subject: [PATCH 02/28] fixed typo based on Julien's comment --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 2969e34..85fe268 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -1,7 +1,7 @@ # Lesion segmentation protocol: The following details the protocol for Multiple Sclerosis (MS) lesion segmentation in the spinal cord. -Imaging the spatial cord is often essential to confirm the diagnosis of MS. That is because the lesions of the spinal cord are included in the McDonald diagnosis criteria, which studies dissemination in space and in time [(Thompson et al. 2018)](https://pubmed.ncbi.nlm.nih.gov/29275977/). While the MAGNIMS-CMSC-NAIMS working group recommends to use at least two sagittal images for MS diagnosis, still, axial imaging is mentioned as optional in international imaging guidelines [(Wattjes et al. 2021)](https://pubmed.ncbi.nlm.nih.gov/34139157/). +Imaging the spinal cord is often essential to confirm the diagnosis of MS. That is because the lesions of the spinal cord are included in the McDonald diagnosis criteria, which studies dissemination in space and in time [(Thompson et al. 2018)](https://pubmed.ncbi.nlm.nih.gov/29275977/). While the MAGNIMS-CMSC-NAIMS working group recommends to use at least two sagittal images for MS diagnosis, still, axial imaging is mentioned as optional in international imaging guidelines [(Wattjes et al. 2021)](https://pubmed.ncbi.nlm.nih.gov/34139157/). For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and STIR contrasts. New studies [(Peters et al. 2024)](https://pubmed.ncbi.nlm.nih.gov/38289376/)[(Fechner et al. 2019)](https://pubmed.ncbi.nlm.nih.gov/30679225/) showed that using PSIR contrasts improved MS lesion detection in the spinal cord. [(Fechner et al. 2019)](https://pubmed.ncbi.nlm.nih.gov/30679225/) showed that the PSIR contrast showed a higher signal-to-noise (SNR) ratio compared to the STIR contrast. ## Criteria to segment MS lesions in the spinal cord: From 6d01e30d8ab971fd5b9bdfca7c7d51916fe8fc88 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Wed, 10 Apr 2024 11:22:23 -0400 Subject: [PATCH 03/28] changed criterias to bullet points --- lesion_segmentation_protocol.md | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 85fe268..9f740d7 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -6,10 +6,10 @@ For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and ## Criteria to segment MS lesions in the spinal cord: -Do not segment lesions in images with too many artifacts (such as this one : https://github.com/ivadomed/canproco/issues/53#issue-1938136790). Preferably, add the image to the exclude file so that it isn’t used for model training… -When segmenting lesions on thick slices, always look at the above/below slices to build the volume of the lesion (this can minimize partial volume effect). -Do not segment lesions above the first vertebrae (because here we focus only on MS lesions in the spinal cord). -For lesions segmentations which you are not 100% sure, flag the subject and report it for external validation of the segmentation. +- Do not segment lesions in images with too many artifacts (such as this one : https://github.com/ivadomed/canproco/issues/53#issue-1938136790). Preferably, add the image to the exclude file so that it isn’t used for model training… +- When segmenting lesions on thick slices, always look at the above/below slices to build the volume of the lesion (this can minimize partial volume effect). +- Do not segment lesions above the first vertebrae (because here we focus only on MS lesions in the spinal cord). +- For lesions segmentations which you are not 100% sure, flag the subject and report it for external validation of the segmentation. ## How to manually segment lesions: From fddd8e3303f93c2276dd42e154ab6992428c8f02 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Wed, 10 Apr 2024 11:23:26 -0400 Subject: [PATCH 04/28] replace link to example by hyperlink --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 9f740d7..f3651ba 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -6,7 +6,7 @@ For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and ## Criteria to segment MS lesions in the spinal cord: -- Do not segment lesions in images with too many artifacts (such as this one : https://github.com/ivadomed/canproco/issues/53#issue-1938136790). Preferably, add the image to the exclude file so that it isn’t used for model training… +- Do not segment lesions in images with too many artifacts (such as this [example](https://github.com/ivadomed/canproco/issues/53#issue-1938136790)). Preferably, add the image to the exclude file so that it isn’t used for model training… - When segmenting lesions on thick slices, always look at the above/below slices to build the volume of the lesion (this can minimize partial volume effect). - Do not segment lesions above the first vertebrae (because here we focus only on MS lesions in the spinal cord). - For lesions segmentations which you are not 100% sure, flag the subject and report it for external validation of the segmentation. From 2e7d7fd1fdd39990d0c16a12e9c6f80688903444 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Wed, 10 Apr 2024 11:28:47 -0400 Subject: [PATCH 05/28] typo correction in introduction Co-authored-by: Maxime B <142258732+maxradx@users.noreply.github.com> --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index f3651ba..f0068c0 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -1,7 +1,7 @@ # Lesion segmentation protocol: The following details the protocol for Multiple Sclerosis (MS) lesion segmentation in the spinal cord. -Imaging the spinal cord is often essential to confirm the diagnosis of MS. That is because the lesions of the spinal cord are included in the McDonald diagnosis criteria, which studies dissemination in space and in time [(Thompson et al. 2018)](https://pubmed.ncbi.nlm.nih.gov/29275977/). While the MAGNIMS-CMSC-NAIMS working group recommends to use at least two sagittal images for MS diagnosis, still, axial imaging is mentioned as optional in international imaging guidelines [(Wattjes et al. 2021)](https://pubmed.ncbi.nlm.nih.gov/34139157/). +Imaging the spinal cord is often essential to confirm the diagnosis of MS. That is because the lesions of the spinal cord are included in the McDonald diagnostic criteria, which considers dissemination in space and in time [(Thompson et al. 2018)](https://pubmed.ncbi.nlm.nih.gov/29275977/). While the MAGNIMS-CMSC-NAIMS working group recommends to use at least two sagittal images for MS diagnosis, still, axial imaging is mentioned as optional in international imaging guidelines [(Wattjes et al. 2021)](https://pubmed.ncbi.nlm.nih.gov/34139157/). For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and STIR contrasts. New studies [(Peters et al. 2024)](https://pubmed.ncbi.nlm.nih.gov/38289376/)[(Fechner et al. 2019)](https://pubmed.ncbi.nlm.nih.gov/30679225/) showed that using PSIR contrasts improved MS lesion detection in the spinal cord. [(Fechner et al. 2019)](https://pubmed.ncbi.nlm.nih.gov/30679225/) showed that the PSIR contrast showed a higher signal-to-noise (SNR) ratio compared to the STIR contrast. ## Criteria to segment MS lesions in the spinal cord: From 3bbcc6ace802daaadf421947c9662eed0b261812 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Wed, 10 Apr 2024 11:32:06 -0400 Subject: [PATCH 06/28] changed json link to json example in the readme --- lesion_segmentation_protocol.md | 15 ++++++++++++++- 1 file changed, 14 insertions(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index f0068c0..6b5a3a2 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -13,7 +13,20 @@ For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and ## How to manually segment lesions: -- MS spinal cord lesions can be manually corrected from the prediction of a segmentation model or manually segmented from scratch. In the first case, make sure to build the json file associated with the segmentation prediction such as detailed in this comment (https://github.com/ivadomed/canproco/issues/73#issuecomment-1912302223). +- MS spinal cord lesions can be manually corrected from the prediction of a segmentation model or manually segmented from scratch. In the first case, make sure to build the json file associated with the segmentation prediction such as the following : + +```json +{ + "GeneratedBy": [ + { + "Name": "2D nnUNet model model_ms_seg_sc-lesion_regionBased.zip", + "Version": "https://github.com/ivadomed/canproco/releases/tag/r20240125", + "Date": "2024-01-26" + } + ] + } +``` + - For manual correction of the segmentation file use the manual-correction (https://github.com/spinalcordtoolbox/manual-correction) repository. The command can be inspired from this: ```console From ffa1b0adbc88ffb6b978634018a2a82c12b7da5e Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Wed, 10 Apr 2024 11:37:35 -0400 Subject: [PATCH 07/28] added examples of complicated lesion segmentations --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 6b5a3a2..afeda66 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -35,7 +35,7 @@ python manual_correction.py -path-img ~/data/canproco -config ~/config_seg.yml - Then a QC should be produced (prefarably using SCT) and added to a Github issue for further validation by other investigators. -- If you are not sure of a subject, it should be flagged on Github for a more open discussion. +- If you are not sure of a subject, it should be flagged on Github for a more open discussion: here are some examples [(1)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947326493) and [(2)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947338624) ## More details: The following section details the different types of errors which occur during lesion segmentation. It is based on the condensed Nascimento Taxonomy: From 61291c714ebe296b6da294a6a9dda6a0b494ea1b Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Wed, 10 Apr 2024 11:44:43 -0400 Subject: [PATCH 08/28] QC explanation reformatting Co-authored-by: Julien Cohen-Adad --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index afeda66..2e717d1 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -33,7 +33,7 @@ For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and python manual_correction.py -path-img ~/data/canproco -config ~/config_seg.yml -path-label ~/data/canproco/derivatives/labels -suffix-files-lesion _lesion-manual -fsleyes-dr="-40,70" ``` -- Then a QC should be produced (prefarably using SCT) and added to a Github issue for further validation by other investigators. +- A Quality Control (QC) report should be produced using SCT, and added to a Github issue for further validation by other investigators. - If you are not sure of a subject, it should be flagged on Github for a more open discussion: here are some examples [(1)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947326493) and [(2)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947338624) From fe015f5268c2ecc70b12879f512e3bb5a349b11b Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 10:14:05 -0400 Subject: [PATCH 09/28] add sct_qc commands --- lesion_segmentation_protocol.md | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 2e717d1..20c6041 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -33,7 +33,11 @@ For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and python manual_correction.py -path-img ~/data/canproco -config ~/config_seg.yml -path-label ~/data/canproco/derivatives/labels -suffix-files-lesion _lesion-manual -fsleyes-dr="-40,70" ``` -- A Quality Control (QC) report should be produced using SCT, and added to a Github issue for further validation by other investigators. +- A Quality Control (QC) report should be produced using SCT, and added to a Github issue for further validation by other investigators. Using SCT, you can review lesion segmentation in the axial or sagittal plane : + +```console +sct_qc -i {image_file} -d {lesion_seg_file} -s {sc_seg_file} -p sct_deepseg_lesion -plane {sagittal, axial} -qc {canproco_qc_folder} +``` - If you are not sure of a subject, it should be flagged on Github for a more open discussion: here are some examples [(1)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947326493) and [(2)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947338624) From 7e18554d89f9b93d8cf1492f35ace97629afc51d Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 10:44:19 -0400 Subject: [PATCH 10/28] added software section --- lesion_segmentation_protocol.md | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 20c6041..085d24b 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -39,7 +39,12 @@ python manual_correction.py -path-img ~/data/canproco -config ~/config_seg.yml sct_qc -i {image_file} -d {lesion_seg_file} -s {sc_seg_file} -p sct_deepseg_lesion -plane {sagittal, axial} -qc {canproco_qc_folder} ``` -- If you are not sure of a subject, it should be flagged on Github for a more open discussion: here are some examples [(1)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947326493) and [(2)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947338624) +- If you are not sure of a subject, it should be flagged on Github for a more open discussion: here are some examples [(1)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947326493) and [(2)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947338624) + +## Step 1: Get familiar with FSLeyes and SCT: +It is common practice to use FSLeyes at NeuroPoly for visual inspection of MRI images and manual segmentation of MS lesions. Therefore, naturally, the first step of the lesion segmentation process is to complete the FSLeyes tutorial ([FSLeyes documentation](https://open.win.ox.ac.uk/pages/fsl/fsleyes/fsleyes/userdoc/) and [video tutorial](https://www.youtube.com/playlist?list=PLIQIswOrUH69qFMNg8KYkEGkvCNEwlnfT)). Trainees are encouraged to learn keyboard shortcuts (ctrl+F to toggle an image, shift+↑ to scroll through volumes, ...). + +Furthermore, it is recommended to get familiar with SCT for creating QCs and for manual correction ([SCT tutorial](https://spinalcordtoolbox.com/user_section/tutorials.html)). ## More details: The following section details the different types of errors which occur during lesion segmentation. It is based on the condensed Nascimento Taxonomy: From b98087bddd45d4e602b3d7a7799bb2281c6203c1 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 10:49:12 -0400 Subject: [PATCH 11/28] added ref at the bottom to other segmentation protocols --- lesion_segmentation_protocol.md | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 085d24b..b4e9337 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -51,3 +51,8 @@ The following section details the different types of errors which occur during l nascimento_taxonomy +## Sources +This lesion segmentation protocol was inspired from these ressources: +- deSouza NM, van der Lugt A, Deroose CM, et al. Standardised lesion segmentation for imaging biomarker quantitation: a consensus recommendation from ESR and EORTC. Insights Imaging. 2022;13(1):159. Published 2022 Oct 4. doi:10.1186/s13244-022-01287-4 : [link](https://pubmed.ncbi.nlm.nih.gov/36194301/) +- Lo BP, Donnelly MR, Barisano G, Liew SL. A standardized protocol for manually segmenting stroke lesions on high-resolution T1-weighted MR images. Front Neuroimaging. 2023;1:1098604. Published 2023 Jan 10. doi:10.3389/fnimg.2022.1098604 : [link](https://pubmed.ncbi.nlm.nih.gov/37555152/) + From 77425fb03bcc37378f49031c92ad790e9b648d25 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 11:12:50 -0400 Subject: [PATCH 12/28] added step 2 : sc anatomy and lesion segmentation --- lesion_segmentation_protocol.md | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index b4e9337..b986b79 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -46,6 +46,15 @@ It is common practice to use FSLeyes at NeuroPoly for visual inspection of MRI i Furthermore, it is recommended to get familiar with SCT for creating QCs and for manual correction ([SCT tutorial](https://spinalcordtoolbox.com/user_section/tutorials.html)). +## Step 2: Spinal cord anatomy and lesion segmentation +Before, moving on to MS lesion segmentation, trainees are advised to study the neuroanatomical structures of healthy spinal cords. Trainees should look at healthy spinal cords in MRI images of different contrasts: T2w, T1w, PSIR, STIR, MP2RAGE... (DISCUSS BUILDING THIS DATASET). + +To learn the specificity of MS lesions, trainees should work on differentiating MS lesions from Spinal Cord Injury (traumatic and non-traumatic) and DCM. A training dataset will be built especially for this case (TO DISCUSS: BUILDING A (PUBLIC) REPO TO TRAIN TO DISTINGUISH PATHOLOGIES). + +One of the most challenging task of MS lesion segmentation is to distinguish the border of a lesion and the cerebrospinal fluid (CSF). To learn where to draw the lesion border, a set of tricky examples validated by a NeuroRadioligist will be created. (TO DISCUSS AS WELL). + +Finally, for trainees will little or no experience with MS lesion segmentation, a checklist will be built to avoid being overwhelmed by the multiple images/contrasts/acquisitions. We typically recommend to start with the view in the highest resolution (often the sagittal view) to first identify lesions, and to move to other contrast/acquisition to validate the segmentation borders and lesion detection. During this step, playing with the brightness and the contrast is key. After locating the lesion to be traced, we recommend starting in a middle slice around the middle of the lesion and then move toward each end of the lesioned area. We also recommended frequently scrolling back and forth around the slice they are tracing on to ensure border consistency. + ## More details: The following section details the different types of errors which occur during lesion segmentation. It is based on the condensed Nascimento Taxonomy: From 6e4a30b6dc887e4a496eaac50787611925b7fb51 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 11:13:19 -0400 Subject: [PATCH 13/28] julien typo fix : remove":" Co-authored-by: Julien Cohen-Adad --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index b986b79..724da35 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -1,4 +1,4 @@ -# Lesion segmentation protocol: +# Lesion segmentation protocol The following details the protocol for Multiple Sclerosis (MS) lesion segmentation in the spinal cord. Imaging the spinal cord is often essential to confirm the diagnosis of MS. That is because the lesions of the spinal cord are included in the McDonald diagnostic criteria, which considers dissemination in space and in time [(Thompson et al. 2018)](https://pubmed.ncbi.nlm.nih.gov/29275977/). While the MAGNIMS-CMSC-NAIMS working group recommends to use at least two sagittal images for MS diagnosis, still, axial imaging is mentioned as optional in international imaging guidelines [(Wattjes et al. 2021)](https://pubmed.ncbi.nlm.nih.gov/34139157/). From 7e0eac3c1580e61939d5559403ccf3152ceb596d Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 11:39:49 -0400 Subject: [PATCH 14/28] typo fix : remove ":" 2 Co-authored-by: Julien Cohen-Adad --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 724da35..021315d 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -11,7 +11,7 @@ For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and - Do not segment lesions above the first vertebrae (because here we focus only on MS lesions in the spinal cord). - For lesions segmentations which you are not 100% sure, flag the subject and report it for external validation of the segmentation. -## How to manually segment lesions: +## How to manually segment lesions - MS spinal cord lesions can be manually corrected from the prediction of a segmentation model or manually segmented from scratch. In the first case, make sure to build the json file associated with the segmentation prediction such as the following : From 359f0519b1924cd9729509533d1c214986a79762 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 11:40:16 -0400 Subject: [PATCH 15/28] rewrite partial volume explanation Co-authored-by: Julien Cohen-Adad --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 021315d..54ccd59 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -7,7 +7,7 @@ For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and ## Criteria to segment MS lesions in the spinal cord: - Do not segment lesions in images with too many artifacts (such as this [example](https://github.com/ivadomed/canproco/issues/53#issue-1938136790)). Preferably, add the image to the exclude file so that it isn’t used for model training… -- When segmenting lesions on thick slices, always look at the above/below slices to build the volume of the lesion (this can minimize partial volume effect). +- When segmenting lesions on thick slices, always look at the adjacent slices, as partial volume effect can sometimes reduce the appearance of a lesion (close to noise level). - Do not segment lesions above the first vertebrae (because here we focus only on MS lesions in the spinal cord). - For lesions segmentations which you are not 100% sure, flag the subject and report it for external validation of the segmentation. From 89283825e21f39c456103e9287c815bb411bf64d Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 11:40:51 -0400 Subject: [PATCH 16/28] rewrite sentence on not segmenting above first vert Co-authored-by: Julien Cohen-Adad --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 54ccd59..956e89b 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -8,7 +8,7 @@ For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and - Do not segment lesions in images with too many artifacts (such as this [example](https://github.com/ivadomed/canproco/issues/53#issue-1938136790)). Preferably, add the image to the exclude file so that it isn’t used for model training… - When segmenting lesions on thick slices, always look at the adjacent slices, as partial volume effect can sometimes reduce the appearance of a lesion (close to noise level). -- Do not segment lesions above the first vertebrae (because here we focus only on MS lesions in the spinal cord). +- Unless otherwise stated, do not segment lesions above the first vertebrae (because here we focus only on MS lesions in the spinal cord). - For lesions segmentations which you are not 100% sure, flag the subject and report it for external validation of the segmentation. ## How to manually segment lesions From f9a370b44cce4f1052317d0e789c079b0efd4832 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 11:41:20 -0400 Subject: [PATCH 17/28] rewrite two options for seg Co-authored-by: Julien Cohen-Adad --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 956e89b..4001e38 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -13,7 +13,7 @@ For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and ## How to manually segment lesions -- MS spinal cord lesions can be manually corrected from the prediction of a segmentation model or manually segmented from scratch. In the first case, make sure to build the json file associated with the segmentation prediction such as the following : +- MS spinal cord lesions can either (i) be automatically segmented from an algorithm and then manually corrected, or (ii) manually segmented from scratch. In the former case, make sure to use the JSON file that was created by the automatic segmentation algorithm, in order to track provenance: ```json { From 11ac1d68b42df84442928edbbc8c355c402b792b Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 11:41:46 -0400 Subject: [PATCH 18/28] rewrite manual correction explanation Co-authored-by: Julien Cohen-Adad --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 4001e38..f887cf5 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -27,7 +27,7 @@ For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and } ``` -- For manual correction of the segmentation file use the manual-correction (https://github.com/spinalcordtoolbox/manual-correction) repository. The command can be inspired from this: +- To manually create/correct the segmentation, please use the manual-correction (https://github.com/spinalcordtoolbox/manual-correction) repository. The command can be inspired from this: ```console python manual_correction.py -path-img ~/data/canproco -config ~/config_seg.yml -path-label ~/data/canproco/derivatives/labels -suffix-files-lesion _lesion-manual -fsleyes-dr="-40,70" From b525c0c1b2c7e485b36b7ec583fb02e37fa35b14 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 11:42:09 -0400 Subject: [PATCH 19/28] rewrite QC Co-authored-by: Julien Cohen-Adad --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index f887cf5..70026e2 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -33,7 +33,7 @@ For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and python manual_correction.py -path-img ~/data/canproco -config ~/config_seg.yml -path-label ~/data/canproco/derivatives/labels -suffix-files-lesion _lesion-manual -fsleyes-dr="-40,70" ``` -- A Quality Control (QC) report should be produced using SCT, and added to a Github issue for further validation by other investigators. Using SCT, you can review lesion segmentation in the axial or sagittal plane : +- A Quality Control (QC) report should be produced using SCT, and added to a GitHub issue for further validation by other investigators. Using SCT, you can review lesion segmentation in the axial or sagittal plane: ```console sct_qc -i {image_file} -d {lesion_seg_file} -s {sc_seg_file} -p sct_deepseg_lesion -plane {sagittal, axial} -qc {canproco_qc_folder} From 5f11e4f98db1ea7ee398d738727a69ef1f056760 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 11:42:46 -0400 Subject: [PATCH 20/28] rewrite protocol when unsure Co-authored-by: Julien Cohen-Adad --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 70026e2..f06b9b9 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -39,7 +39,7 @@ python manual_correction.py -path-img ~/data/canproco -config ~/config_seg.yml sct_qc -i {image_file} -d {lesion_seg_file} -s {sc_seg_file} -p sct_deepseg_lesion -plane {sagittal, axial} -qc {canproco_qc_folder} ``` -- If you are not sure of a subject, it should be flagged on Github for a more open discussion: here are some examples [(1)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947326493) and [(2)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947338624) +- If you are not sure about the segmentation on a subject, it should be flagged on GitHub for a more open discussion: here are some examples [(1)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947326493) and [(2)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947338624) ## Step 1: Get familiar with FSLeyes and SCT: It is common practice to use FSLeyes at NeuroPoly for visual inspection of MRI images and manual segmentation of MS lesions. Therefore, naturally, the first step of the lesion segmentation process is to complete the FSLeyes tutorial ([FSLeyes documentation](https://open.win.ox.ac.uk/pages/fsl/fsleyes/fsleyes/userdoc/) and [video tutorial](https://www.youtube.com/playlist?list=PLIQIswOrUH69qFMNg8KYkEGkvCNEwlnfT)). Trainees are encouraged to learn keyboard shortcuts (ctrl+F to toggle an image, shift+↑ to scroll through volumes, ...). From 29829823f2db883a273b5b40ff66770c48aa5a40 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 11:44:02 -0400 Subject: [PATCH 21/28] change last section name Co-authored-by: Julien Cohen-Adad --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index f06b9b9..27688b9 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -55,7 +55,7 @@ One of the most challenging task of MS lesion segmentation is to distinguish the Finally, for trainees will little or no experience with MS lesion segmentation, a checklist will be built to avoid being overwhelmed by the multiple images/contrasts/acquisitions. We typically recommend to start with the view in the highest resolution (often the sagittal view) to first identify lesions, and to move to other contrast/acquisition to validate the segmentation borders and lesion detection. During this step, playing with the brightness and the contrast is key. After locating the lesion to be traced, we recommend starting in a middle slice around the middle of the lesion and then move toward each end of the lesioned area. We also recommended frequently scrolling back and forth around the slice they are tracing on to ensure border consistency. -## More details: +## Taxonomy to evaluation lesion segmentation The following section details the different types of errors which occur during lesion segmentation. It is based on the condensed Nascimento Taxonomy: nascimento_taxonomy From e403321732ef53b4d64bf7201f996fa5e6e1b0ab Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 11:55:49 -0400 Subject: [PATCH 22/28] update dataset of multiple contrasts --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 27688b9..902e130 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -47,7 +47,7 @@ It is common practice to use FSLeyes at NeuroPoly for visual inspection of MRI i Furthermore, it is recommended to get familiar with SCT for creating QCs and for manual correction ([SCT tutorial](https://spinalcordtoolbox.com/user_section/tutorials.html)). ## Step 2: Spinal cord anatomy and lesion segmentation -Before, moving on to MS lesion segmentation, trainees are advised to study the neuroanatomical structures of healthy spinal cords. Trainees should look at healthy spinal cords in MRI images of different contrasts: T2w, T1w, PSIR, STIR, MP2RAGE... (DISCUSS BUILDING THIS DATASET). +Before, moving on to MS lesion segmentation, trainees are advised to study the neuroanatomical structures of healthy spinal cords. Trainees should look at healthy spinal cords in MRI images of different contrasts: T2w, T1w, PSIR, STIR, MP2RAGE... A public dataset will be built for this purpose. To learn the specificity of MS lesions, trainees should work on differentiating MS lesions from Spinal Cord Injury (traumatic and non-traumatic) and DCM. A training dataset will be built especially for this case (TO DISCUSS: BUILDING A (PUBLIC) REPO TO TRAIN TO DISTINGUISH PATHOLOGIES). From be196413b46a48b6a8f37a89660e629e1108e883 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 11:58:27 -0400 Subject: [PATCH 23/28] remove dcm and sci --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 902e130..1fa4cce 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -49,7 +49,7 @@ Furthermore, it is recommended to get familiar with SCT for creating QCs and for ## Step 2: Spinal cord anatomy and lesion segmentation Before, moving on to MS lesion segmentation, trainees are advised to study the neuroanatomical structures of healthy spinal cords. Trainees should look at healthy spinal cords in MRI images of different contrasts: T2w, T1w, PSIR, STIR, MP2RAGE... A public dataset will be built for this purpose. -To learn the specificity of MS lesions, trainees should work on differentiating MS lesions from Spinal Cord Injury (traumatic and non-traumatic) and DCM. A training dataset will be built especially for this case (TO DISCUSS: BUILDING A (PUBLIC) REPO TO TRAIN TO DISTINGUISH PATHOLOGIES). +To learn the specificity of MS lesions, trainees should work on differentiating MS lesions from other diseases. One of the most challenging task of MS lesion segmentation is to distinguish the border of a lesion and the cerebrospinal fluid (CSF). To learn where to draw the lesion border, a set of tricky examples validated by a NeuroRadioligist will be created. (TO DISCUSS AS WELL). From a0219960b1ad18714b8138f101927785c6b2c871 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 12:00:40 -0400 Subject: [PATCH 24/28] remove comment --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 1fa4cce..c4774a6 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -51,7 +51,7 @@ Before, moving on to MS lesion segmentation, trainees are advised to study the n To learn the specificity of MS lesions, trainees should work on differentiating MS lesions from other diseases. -One of the most challenging task of MS lesion segmentation is to distinguish the border of a lesion and the cerebrospinal fluid (CSF). To learn where to draw the lesion border, a set of tricky examples validated by a NeuroRadioligist will be created. (TO DISCUSS AS WELL). +One of the most challenging task of MS lesion segmentation is to distinguish the border of a lesion and the cerebrospinal fluid (CSF). To learn where to draw the lesion border, a set of tricky examples validated by a NeuroRadioligist will be created. Finally, for trainees will little or no experience with MS lesion segmentation, a checklist will be built to avoid being overwhelmed by the multiple images/contrasts/acquisitions. We typically recommend to start with the view in the highest resolution (often the sagittal view) to first identify lesions, and to move to other contrast/acquisition to validate the segmentation borders and lesion detection. During this step, playing with the brightness and the contrast is key. After locating the lesion to be traced, we recommend starting in a middle slice around the middle of the lesion and then move toward each end of the lesioned area. We also recommended frequently scrolling back and forth around the slice they are tracing on to ensure border consistency. From 657b9bb34d46f7ab84398abf7628de27f48f5404 Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 12:11:00 -0400 Subject: [PATCH 25/28] add step 3 --- lesion_segmentation_protocol.md | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 deletions(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index c4774a6..1c95817 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -9,7 +9,7 @@ For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and - Do not segment lesions in images with too many artifacts (such as this [example](https://github.com/ivadomed/canproco/issues/53#issue-1938136790)). Preferably, add the image to the exclude file so that it isn’t used for model training… - When segmenting lesions on thick slices, always look at the adjacent slices, as partial volume effect can sometimes reduce the appearance of a lesion (close to noise level). - Unless otherwise stated, do not segment lesions above the first vertebrae (because here we focus only on MS lesions in the spinal cord). -- For lesions segmentations which you are not 100% sure, flag the subject and report it for external validation of the segmentation. +- If you have any doubt and/or are not 100% confident about one (or more) lesion(s) segmentation(s), flag the subject and report it for external validation of the segmentation. ## How to manually segment lesions @@ -27,7 +27,7 @@ For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and } ``` -- To manually create/correct the segmentation, please use the manual-correction (https://github.com/spinalcordtoolbox/manual-correction) repository. The command can be inspired from this: +- To manually create/correct the segmentation, please use the manual-correction (https://github.com/spinalcordtoolbox/manual-correction) repository. The command can be inspired by this: ```console python manual_correction.py -path-img ~/data/canproco -config ~/config_seg.yml -path-label ~/data/canproco/derivatives/labels -suffix-files-lesion _lesion-manual -fsleyes-dr="-40,70" @@ -51,17 +51,20 @@ Before, moving on to MS lesion segmentation, trainees are advised to study the n To learn the specificity of MS lesions, trainees should work on differentiating MS lesions from other diseases. -One of the most challenging task of MS lesion segmentation is to distinguish the border of a lesion and the cerebrospinal fluid (CSF). To learn where to draw the lesion border, a set of tricky examples validated by a NeuroRadioligist will be created. +One of the most challenging tasks of MS lesion segmentation is to distinguish the border of a lesion and the cerebrospinal fluid (CSF). To learn where to draw the lesion border, a set of tricky examples validated by a NeuroRadioligist will be created. -Finally, for trainees will little or no experience with MS lesion segmentation, a checklist will be built to avoid being overwhelmed by the multiple images/contrasts/acquisitions. We typically recommend to start with the view in the highest resolution (often the sagittal view) to first identify lesions, and to move to other contrast/acquisition to validate the segmentation borders and lesion detection. During this step, playing with the brightness and the contrast is key. After locating the lesion to be traced, we recommend starting in a middle slice around the middle of the lesion and then move toward each end of the lesioned area. We also recommended frequently scrolling back and forth around the slice they are tracing on to ensure border consistency. +Finally, for trainees will little or no experience with MS lesion segmentation, a checklist will be built to avoid being overwhelmed by the multiple images/contrasts/acquisitions. We typically recommend starting with the view in the highest resolution (often the sagittal view) to first identify lesions and to move to other contrast/acquisition to validate the segmentation borders and lesion detection. During this step, playing with the brightness and the contrast is key. After locating the lesion to be traced, we recommend starting in a middle slice around the middle of the lesion and then moving toward each end of the lesioned area. We also recommended frequently scrolling back and forth around the slice they are tracing on to ensure border consistency. -## Taxonomy to evaluation lesion segmentation +## Step 3: Manual segmentation assessment +After manual segmentation of MS SC lesion in 5 cases, trainees will receive feedback (from a NeuroRadiologist or a comparison with a QC with the real segmentation). One week later, they will be asked to re-segment the same images as well as 2 other images without using their previous segmentation to validate their improvements. The segmentation should also be accompanied by a JSON file for the data to be BIDS compliant. + +## Taxonomy to evaluate lesion segmentation The following section details the different types of errors which occur during lesion segmentation. It is based on the condensed Nascimento Taxonomy: nascimento_taxonomy ## Sources -This lesion segmentation protocol was inspired from these ressources: +This lesion segmentation protocol was inspired by these resources: - deSouza NM, van der Lugt A, Deroose CM, et al. Standardised lesion segmentation for imaging biomarker quantitation: a consensus recommendation from ESR and EORTC. Insights Imaging. 2022;13(1):159. Published 2022 Oct 4. doi:10.1186/s13244-022-01287-4 : [link](https://pubmed.ncbi.nlm.nih.gov/36194301/) - Lo BP, Donnelly MR, Barisano G, Liew SL. A standardized protocol for manually segmenting stroke lesions on high-resolution T1-weighted MR images. Front Neuroimaging. 2023;1:1098604. Published 2023 Jan 10. doi:10.3389/fnimg.2022.1098604 : [link](https://pubmed.ncbi.nlm.nih.gov/37555152/) From 63f305ca019cdcb6cb92341a695d1ba7e55412cf Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 12:12:13 -0400 Subject: [PATCH 26/28] lesion definition suggestion Co-authored-by: Maxime B <142258732+maxradx@users.noreply.github.com> --- lesion_segmentation_protocol.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 1c95817..63fdd69 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -5,6 +5,8 @@ Imaging the spinal cord is often essential to confirm the diagnosis of MS. That For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and STIR contrasts. New studies [(Peters et al. 2024)](https://pubmed.ncbi.nlm.nih.gov/38289376/)[(Fechner et al. 2019)](https://pubmed.ncbi.nlm.nih.gov/30679225/) showed that using PSIR contrasts improved MS lesion detection in the spinal cord. [(Fechner et al. 2019)](https://pubmed.ncbi.nlm.nih.gov/30679225/) showed that the PSIR contrast showed a higher signal-to-noise (SNR) ratio compared to the STIR contrast. ## Criteria to segment MS lesions in the spinal cord: +Based on the definition proposed for MS lesions in the spinal cord in the 2017 revisions of the McDonald criteria: +"A hyperintense lesion in the cervical, thoracic, or lumbar spinal cord seen on T2 plus short tau inversion recovery, proton-density images, or other appropriate sequences, or in two planes on T2 images." [(Thompson et al. 2018)](https://pubmed.ncbi.nlm.nih.gov/29275977/) [(Filippi et al. 2016)](https://pubmed.ncbi.nlm.nih.gov/26822746/) [(Brownlee et al. 2018)](https://pubmed.ncbi.nlm.nih.gov/27889190/) [(Rovira et al. 2016)](https://pubmed.ncbi.nlm.nih.gov/26149978/) - Do not segment lesions in images with too many artifacts (such as this [example](https://github.com/ivadomed/canproco/issues/53#issue-1938136790)). Preferably, add the image to the exclude file so that it isn’t used for model training… - When segmenting lesions on thick slices, always look at the adjacent slices, as partial volume effect can sometimes reduce the appearance of a lesion (close to noise level). From 13ca69266c8fb5132f9475005da33bccd0de70fa Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 12:12:31 -0400 Subject: [PATCH 27/28] typo fix --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 63fdd69..4c30cdd 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -4,7 +4,7 @@ The following details the protocol for Multiple Sclerosis (MS) lesion segmentati Imaging the spinal cord is often essential to confirm the diagnosis of MS. That is because the lesions of the spinal cord are included in the McDonald diagnostic criteria, which considers dissemination in space and in time [(Thompson et al. 2018)](https://pubmed.ncbi.nlm.nih.gov/29275977/). While the MAGNIMS-CMSC-NAIMS working group recommends to use at least two sagittal images for MS diagnosis, still, axial imaging is mentioned as optional in international imaging guidelines [(Wattjes et al. 2021)](https://pubmed.ncbi.nlm.nih.gov/34139157/). For detecting MS lesions in the spinal cord, two main contrasts emerge: PSIR and STIR contrasts. New studies [(Peters et al. 2024)](https://pubmed.ncbi.nlm.nih.gov/38289376/)[(Fechner et al. 2019)](https://pubmed.ncbi.nlm.nih.gov/30679225/) showed that using PSIR contrasts improved MS lesion detection in the spinal cord. [(Fechner et al. 2019)](https://pubmed.ncbi.nlm.nih.gov/30679225/) showed that the PSIR contrast showed a higher signal-to-noise (SNR) ratio compared to the STIR contrast. -## Criteria to segment MS lesions in the spinal cord: +## Criteria to segment MS lesions in the spinal cord Based on the definition proposed for MS lesions in the spinal cord in the 2017 revisions of the McDonald criteria: "A hyperintense lesion in the cervical, thoracic, or lumbar spinal cord seen on T2 plus short tau inversion recovery, proton-density images, or other appropriate sequences, or in two planes on T2 images." [(Thompson et al. 2018)](https://pubmed.ncbi.nlm.nih.gov/29275977/) [(Filippi et al. 2016)](https://pubmed.ncbi.nlm.nih.gov/26822746/) [(Brownlee et al. 2018)](https://pubmed.ncbi.nlm.nih.gov/27889190/) [(Rovira et al. 2016)](https://pubmed.ncbi.nlm.nih.gov/26149978/) From 888513d99da6f1ccab2fe0115caf6d9eb14b0bad Mon Sep 17 00:00:00 2001 From: Pierre-Louis Benveniste <67429280+plbenveniste@users.noreply.github.com> Date: Thu, 11 Apr 2024 12:14:29 -0400 Subject: [PATCH 28/28] changed step 1 title --- lesion_segmentation_protocol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lesion_segmentation_protocol.md b/lesion_segmentation_protocol.md index 4c30cdd..57c5fc3 100644 --- a/lesion_segmentation_protocol.md +++ b/lesion_segmentation_protocol.md @@ -43,7 +43,7 @@ sct_qc -i {image_file} -d {lesion_seg_file} -s {sc_seg_file} -p sct_deepseg_lesi - If you are not sure about the segmentation on a subject, it should be flagged on GitHub for a more open discussion: here are some examples [(1)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947326493) and [(2)](https://github.com/ivadomed/ms-lesion-agnostic/issues/4#issuecomment-1947338624) -## Step 1: Get familiar with FSLeyes and SCT: +## Step 1: Software to create/correct lesions It is common practice to use FSLeyes at NeuroPoly for visual inspection of MRI images and manual segmentation of MS lesions. Therefore, naturally, the first step of the lesion segmentation process is to complete the FSLeyes tutorial ([FSLeyes documentation](https://open.win.ox.ac.uk/pages/fsl/fsleyes/fsleyes/userdoc/) and [video tutorial](https://www.youtube.com/playlist?list=PLIQIswOrUH69qFMNg8KYkEGkvCNEwlnfT)). Trainees are encouraged to learn keyboard shortcuts (ctrl+F to toggle an image, shift+↑ to scroll through volumes, ...). Furthermore, it is recommended to get familiar with SCT for creating QCs and for manual correction ([SCT tutorial](https://spinalcordtoolbox.com/user_section/tutorials.html)).