ESMRMB Wiki - User contributions [en]

Self-learning MR

2025-09-29T12:50:21Z

MoritzZaiss:

[[Category:Accepted_MS_2025]]



{{Member Session
| title= Self-learning MR
| short_text= Sequence Discovery & Optimization - where we come from, where we go.
| duration= 60m - 90m
| audience= Physicists, Radiologists, AI-folks
| Moderators= Moritz Zaiss, Shaihan Malik
| Moderators_Confirmed= Zaiss, Malik
| Speaker1= Tony Stöcker
| Speaker1_Confirmed= yes
| Speaker2= Or Perlman
| Speaker2_Confirmed= yes
|Speaker3=Moritz Zaiss|Speaker4=Shaihan Malik|Speaker3_Confirmed=Yes|Speaker4_Confirmed=Yes}}



== Session description ==

In recent years, learning-based reconstruction and post-processing have transformed clinical MRI—making scans faster, sharper, and more informative. However, as data processing continues to improve, the focus is shifting back to data acquisition. With smaller artifacts becoming more apparent, we need to fix them again in acquisition. With reconstruction methods extracting all available information, the frontier of generation of new contrast features gets a revival.

This session addresses all kind of automized sequence discovery or optimization approaches - from historic EPG-based flip angle train optimization, to MRF schedules - trying to put it under the name "Self-learning MR". Yet, self-learning MR has many names: Optimal Control, End-to-End Optimization, Known-Operator Learning, Physics-Informed AI, Strategy Discovery, Optimal Design of Experiments, Dynamic Inverse Problems

While learning-based methods are well-known on the reconstruction side, we explicitly focus here on the MR preparation/acquisition side in the broadest sense. Of course, some tailored acquisitions can go hand in hand with tailored reconstruction.

All materials are shared via github: https://github.com/mritogether/ESMRMB2025_Self_learning_MR

The program will be the following:

=== Program for first hour: 4 talks ===

* Welcome and Definition of the session (Focus on Acquisition, beyond k-space patterns and recon) (Shaihan and Moritz) 2 mins
* '''Tony Stöcker, DZNE, Bonn: “EPG-based signal modeling and sequence optimization”''' (15 min)
* '''Shaihan Malik, KCL, London: “Parallel Transmit, from fields to excitations to sequences”''' (15 min)
* '''Moritz Zaiss, FAU, Erlangen: “End-to-end learning with MR-zero”''' (15 min)
* '''Or Perlman, Tel Aviv University, “Towards Automatically Optimized Multi-Metabolite CEST Fingerprinting – When a Computational Graph Meets Proton Exchange"''' (15 min)

=== Program for the second half an hour: demos of open source “self-learning” tools ===

* '''Svenja Niesen "Optimize it! EPGs in Python"''' (5 min) Paper: Availabe soon. Code:<nowiki>https://github.com/mrphysics-bonn/epg-tutorial</nowiki>
* '''Jonathan Endres – Pulseq-zero''' (5 min): Paper: <nowiki>https://github.com/pulseq-frame/pulseq-zero/blob/main/abstract/abstract.md</nowiki> Code: <nowiki>https://github.com/pulseq-frame/pulseq-zero</nowiki> DSC + DESC
* '''Felix Glang – Trajectory Optimization''' (5 min):

Paper: <nowiki>https://archive.ismrm.org/2021/4200.html</nowiki> Code:

* '''Dario Bosch – fast pTX''' (5 min) Paper: <nowiki>https://doi.org/10.1007/s10334-023-01134-7</nowiki> Code: <nowiki>https://github.com/dabosch/FastPtx</nowiki>
* '''Nikita Vladimirov: – CEST MRF for everyone''' (5 min) Paper: <nowiki>https://doi.org/10.1038/s41596-025-01152-w</nowiki> Code: <nowiki>https://github.com/momentum-laboratory/molecular-mrf</nowiki>
* Virtual:
** '''Jannik Stebani/Martin Blaimer - Optimization of quantification using Cramer-Rao Lower bound''' (5 min) Paper: Code:<nowiki>https://github.com/HarmonizedMRI/OpenMRF</nowiki>

Self-learning MR

2025-09-29T12:49:34Z

MoritzZaiss:

[[Category:Accepted_MS_2025]]



{{Member Session
| title= Self-learning MR
| short_text= Sequence Discovery & Optimization - where we come from, where we go.
| duration= 60m - 90m
| audience= Physicists, Radiologists, AI-folks
| Moderators= Moritz Zaiss, Shaihan Malik
| Moderators_Confirmed= Zaiss, Malik
| Speaker1= Tony Stöcker
| Speaker1_Confirmed= yes
| Speaker2= Or Perlman
| Speaker2_Confirmed= yes
|Speaker3=Moritz Zaiss|Speaker4=Shaihan Malik}}



== Session description ==

In recent years, learning-based reconstruction and post-processing have transformed clinical MRI—making scans faster, sharper, and more informative. However, as data processing continues to improve, the focus is shifting back to data acquisition. With smaller artifacts becoming more apparent, we need to fix them again in acquisition. With reconstruction methods extracting all available information, the frontier of generation of new contrast features gets a revival.

This session addresses all kind of automized sequence discovery or optimization approaches - from historic EPG-based flip angle train optimization, to MRF schedules - trying to put it under the name "Self-learning MR". Yet, self-learning MR has many names: Optimal Control, End-to-End Optimization, Known-Operator Learning, Physics-Informed AI, Strategy Discovery, Optimal Design of Experiments, Dynamic Inverse Problems

While learning-based methods are well-known on the reconstruction side, we explicitly focus here on the MR preparation/acquisition side in the broadest sense. Of course, some tailored acquisitions can go hand in hand with tailored reconstruction.

All materials are shared via github: https://github.com/mritogether/ESMRMB2025_Self_learning_MR

The program will be the following:

=== Program for first hour: 4 talks ===

* Welcome and Definition of the session (Focus on Acquisition, beyond k-space patterns and recon) (Shaihan and Moritz) 2 mins
* '''Tony Stöcker, DZNE, Bonn: “EPG-based signal modeling and sequence optimization”''' (15 min)
* '''Shaihan Malik, KCL, London: “Parallel Transmit, from fields to excitations to sequences”''' (15 min)
* '''Moritz Zaiss, FAU, Erlangen: “End-to-end learning with MR-zero”''' (15 min)
* '''Or Perlman, Tel Aviv University, “Towards Automatically Optimized Multi-Metabolite CEST Fingerprinting – When a Computational Graph Meets Proton Exchange"''' (15 min)

=== Program for the second half an hour: demos of open source “self-learning” tools ===

* '''Svenja Niesen "Optimize it! EPGs in Python"''' (5 min) Paper: Availabe soon. Code:<nowiki>https://github.com/mrphysics-bonn/epg-tutorial</nowiki>
* '''Jonathan Endres – Pulseq-zero''' (5 min): Paper: <nowiki>https://github.com/pulseq-frame/pulseq-zero/blob/main/abstract/abstract.md</nowiki> Code: <nowiki>https://github.com/pulseq-frame/pulseq-zero</nowiki> DSC + DESC
* '''Felix Glang – Trajectory Optimization''' (5 min):

Paper: <nowiki>https://archive.ismrm.org/2021/4200.html</nowiki> Code:

* '''Dario Bosch – fast pTX''' (5 min) Paper: <nowiki>https://doi.org/10.1007/s10334-023-01134-7</nowiki> Code: <nowiki>https://github.com/dabosch/FastPtx</nowiki>
* '''Nikita Vladimirov: – CEST MRF for everyone''' (5 min) Paper: <nowiki>https://doi.org/10.1038/s41596-025-01152-w</nowiki> Code: <nowiki>https://github.com/momentum-laboratory/molecular-mrf</nowiki>
* Virtual:
** '''Jannik Stebani/Martin Blaimer - Optimization of quantification using Cramer-Rao Lower bound''' (5 min) Paper: Code:<nowiki>https://github.com/HarmonizedMRI/OpenMRF</nowiki>

Self-learning MR

2025-09-29T12:40:12Z

MoritzZaiss: /* Session description */

[[Category:Accepted_MS_2025]]



{{Member Session
| title= Self-learning MR
| short_text= Sequence Discovery & Optimization - where we come from, where we go.
| duration= 60m - 90m
| audience= Physicists, Radiologists, AI-folks
| Moderators= Moritz Zaiss, Shaihan Malik
| Moderators_Confirmed= Zaiss, Malik
| Speaker1= Tony Stöcker
| Speaker1_Confirmed= yes
| Speaker2= Or Perlman
| Speaker2_Confirmed= yes
}}



== Session description ==

In recent years, learning-based reconstruction and post-processing have transformed clinical MRI—making scans faster, sharper, and more informative. However, as data processing continues to improve, the focus is shifting back to data acquisition. With smaller artifacts becoming more apparent, we need to fix them again in acquisition. With reconstruction methods extracting all available information, the frontier of generation of new contrast features gets a revival.

This session addresses all kind of automized sequence discovery or optimization approaches - from historic EPG-based flip angle train optimization, to MRF schedules - trying to put it under the name "Self-learning MR". Yet, self-learning MR has many names: Optimal Control, End-to-End Optimization, Known-Operator Learning, Physics-Informed AI, Strategy Discovery, Optimal Design of Experiments, Dynamic Inverse Problems

While learning-based methods are well-known on the reconstruction side, we explicitly focus here on the MR preparation/acquisition side in the broadest sense. Of course, some tailored acquisitions can go hand in hand with tailored reconstruction.

All materials are shared via github: https://github.com/mritogether/ESMRMB2025_Self_learning_MR

The program will be the following:

=== Program for first hour: 4 talks ===

* Welcome and Definition of the session (Focus on Acquisition, beyond k-space patterns and recon) (Shaihan and Moritz) 2 mins
* '''Tony Stöcker, DZNE, Bonn: “EPG-based signal modeling and sequence optimization”''' (15 min)
* '''Shaihan Malik, KCL, London: “Parallel Transmit, from fields to excitations to sequences”''' (15 min)
* '''Moritz Zaiss, FAU, Erlangen: “End-to-end learning with MR-zero”''' (15 min)
* '''Or Perlman, Tel Aviv University, “Towards Automatically Optimized Multi-Metabolite CEST Fingerprinting – When a Computational Graph Meets Proton Exchange"''' (15 min)

=== Program for the second half an hour: 4-6 demos of open source “self-learning” tools ===

* '''Svenja Niesen "Optimize it! EPGs in Python"''' (5 min) Paper: Availabe soon. Code:<nowiki>https://github.com/mrphysics-bonn/epg-tutorial</nowiki>
* '''Jonathan Endres – Pulseq-zero''' (5 min): Paper: <nowiki>https://github.com/pulseq-frame/pulseq-zero/blob/main/abstract/abstract.md</nowiki> Code: <nowiki>https://github.com/pulseq-frame/pulseq-zero</nowiki> DSC + DESC
* '''Felix Glang – Trajectory Optimization''' (5 min):

Paper: <nowiki>https://archive.ismrm.org/2021/4200.html</nowiki> Code:

* '''Dario Bosch – fast pTX''' (5 min) Paper: <nowiki>https://doi.org/10.1007/s10334-023-01134-7</nowiki> Code: <nowiki>https://github.com/dabosch/FastPtx</nowiki>
* '''Nikita Vladimirov: – CEST MRF for everyone''' (5 min) Paper: <nowiki>https://doi.org/10.1038/s41596-025-01152-w</nowiki> Code: <nowiki>https://github.com/momentum-laboratory/molecular-mrf</nowiki>

Data crime detectives

2025-02-24T08:58:19Z

MoritzZaiss: speakers confirmed

{{Member Session
| title= Data crime detectives
| short_text= A session about data crimes in MR recon
| duration= 60m
| audience= Physicists, Radiologists
| Moderators= Moritz Zaiss, Martin Uecker
| Moderators_Confirmed= MZ confirmed
| Speaker1= Reinhard Heckel (15 min)
| Speaker1_Confirmed= Yes
| Speaker2= Efrat Shimron (15 min)
|Speaker2_Confirmed=No|Speaker3=Moritz Zaiss|Speaker3_confirmed=yes}}



== Session description ==

This session will be about Data Crimes in MRI reconstructon and how simmulation can help.

Retrospective subsampling, commonly used in developing reconstruction algorithms, introduces challenges that can bias results and hinder reproducibility. When k-space data synthesized from reconstructed images, such as those from public repositories, are subsampled, hidden preprocessing pipelines (e.g., lossy compression) often enhance data sparsity. This artificially boosts algorithm performance during evaluation, leading to overly optimistic results that fail to generalize to raw data (https://doi.org/10.1073/pnas.2117203119). Another challenge emerges in transient sequences, such as MP-RAGE or FLAIR, where signal evolution depends on the history of k-space acquisitions itself. As demonstrated in recent studies (https://doi.org/10.1007/978-3-031-16446-0_67, https://doi.org/10.1007/978-3-031-72104-5_23), retrospective subsampling of these sequences fails to account for altered transient magnetization dynamics. For example, subsampling k-space lines effectively removes excitation or refocusing pulses from the prospective sequence, leading to k-space data corresponding to sharper reconstructed images than those of the original sequence. If not performed with care, this approach can cause data domain shifts or data distribution shifts (https://arxiv.org/pdf/2204.07204) between retrospectively subsampled and prospectively acquired data. To raise awareness of such "data crimes" we want to dedicate this session playing detective in public datasets, papers and hands-on simulations.

The program will be the following:

* Introduction ''(10 min)''
* Presentation by Reinhard Heckel ''(15 min) (Confirmed:Yes)''
* Presentation by Efrat Shimron ''(15 min) (Confirmed: Virtual Yes)''
* Presentation by Moritz Zaiss on Simulations ''(15 min) (Confirmed:Yes)''
* Questions and Answers ''(15 min)''

Data crime detectives

2025-02-24T08:57:56Z

MoritzZaiss: confirmed speaker

{{Member Session
| title= Data crime detectives
| short_text= A session about data crimes in MR recon
| duration= 60m
| audience= Physicists, Radiologists
| Moderators= Moritz Zaiss, Martin Uecker
| Moderators_Confirmed= MZ confirmed
| Speaker1= Reinhard Heckel (15 min)
| Speaker1_Confirmed= Yes
| Speaker2= Efrat Shimron (15 min)
|Speaker2_Confirmed=No|Speaker3=Moritz Zaiss|Speaker3_confirmed=yes}}



== Session description ==

This session will be about Data Crimes in MRI reconstructon and how simmulation can help.

Retrospective subsampling, commonly used in developing reconstruction algorithms, introduces challenges that can bias results and hinder reproducibility. When k-space data synthesized from reconstructed images, such as those from public repositories, are subsampled, hidden preprocessing pipelines (e.g., lossy compression) often enhance data sparsity. This artificially boosts algorithm performance during evaluation, leading to overly optimistic results that fail to generalize to raw data (https://doi.org/10.1073/pnas.2117203119). Another challenge emerges in transient sequences, such as MP-RAGE or FLAIR, where signal evolution depends on the history of k-space acquisitions itself. As demonstrated in recent studies (https://doi.org/10.1007/978-3-031-16446-0_67, https://doi.org/10.1007/978-3-031-72104-5_23), retrospective subsampling of these sequences fails to account for altered transient magnetization dynamics. For example, subsampling k-space lines effectively removes excitation or refocusing pulses from the prospective sequence, leading to k-space data corresponding to sharper reconstructed images than those of the original sequence. If not performed with care, this approach can cause data domain shifts or data distribution shifts (https://arxiv.org/pdf/2204.07204) between retrospectively subsampled and prospectively acquired data. To raise awareness of such "data crimes" we want to dedicate this session playing detective in public datasets, papers and hands-on simulations.

The program will be the following:

* Introduction ''(10 min)''
* Presentation by Reinhard Heckel ''(15 min) (Confirmed:No)''
* Presentation by Efrat Shimron ''(15 min) (Confirmed: Virtual Yes)''
* Presentation by Moritz Zaiss on Simulations ''(15 min) (Confirmed:Yes)''
* Questions and Answers ''(15 min)''

Self-learning MR

2025-02-24T08:47:27Z

MoritzZaiss: speakers confirmed

{{Member Session
| title= Self-learning MR
| short_text= Sequence Discovery & Optimization - where we come from, where we go.
| duration= 60m - 90m
| audience= Physicists, Radiologists, AI-folks
| Moderators= Moritz Zaiss, Shaihan Malik
| Moderators_Confirmed= Zaiss, Malik
| Speaker1= Tony Stöcker
| Speaker1_Confirmed= yes
| Speaker2= Or Perlman
| Speaker2_Confirmed= yes
}}



== Session description ==

In recent years, learning-based reconstruction and post-processing have transformed clinical MRI—making scans faster, sharper, and more informative. However, as data processing continues to improve, the focus is shifting back to data acquisition. With smaller artifacts becoming more apparent, we need to fix them again in acquisition. With reconstruction methods extracting all available information, the frontier of generation of new contrast features gets a revival.

This session addresses all kind of automized sequence discovery or optimization approaches - from historic EPG-based flip angle train optimization, to pulse design, and MRF schedules - trying to put it under the name "Self-learning MR". Yet, self-learning MR has many names: Optimal Control, End-to-End Optimization, Known-Operator Learning, Physics-Informed AI, Strategy Discovery, Optimal Design of Experiments, Dynamic Inverse Problems

Potential topics could be:
Self-learning MRF / CEST MRF / MRS
Self-learning trajectories
Self-learning contrast, sharpness
Self-learning over a real scanner
Self-learning imperfection mitigation
Self-learning pulses, pTX/SMS pulses
Self-learning flip angle trains, phases

While learning-based methods are well-known on the reconstruction side, we explicitly focus here on the MR preparation/acquisition side in the broadest sense. Of course, some tailored acquisitions can go hand in hand with tailored reconstruction.

Speakers:
Tony Stöcker - classic EPG-based optimizations of TSE, SFFP. pTX pulse design.
Or Perlman - Overview on MRF, autoseq, and autoCEST.
Moritz Zaiss – End-to-end sequence design

We already have an interest group mailing list with 19 groups/people, who I could invite to submit an abstract if this session gets chosen.

The program will be the following:

Option 1: (1:30h)

*Presentation by Tony Stöcker on history of EPG-based optimization + advances + pTx ''(20 min)''
* Presentation by Moritz Zaiss on end to end learing ''(10 min)''
* Presentation by Or Perlmann on AutoCEST''(10 min)''
* Selected ESMRMB Abstracts that fit to the topic: ''(50 min)''

Option 2: (1h)

* Presentation by Tony Stöcker on history of EPG-baased optimization + advances ''(15 min)''
* Presentation by N. Boulant or Tony Stöcker on Ptx pulses ''(10 min)''
* Presentation by Moritz Zaiss on end to end learing ''(15 min)''
* Presentation by Or Perlmann on AutoCEST''(15 min)''
* Discussion ''(5 min)''

Self-learning MR

2025-02-21T17:08:47Z

MoritzZaiss: duration

{{Member Session
| title= Self-learning MR
| short_text= Sequence Discovery & Optimization - where we come from, where we go.
| duration= 60m - 90m
| audience= Physicists, Radiologists, AI-folks
| Moderators= Moritz Zaiss, Shaihan Malik
| Moderators_Confirmed= Zaiss, Malik
| Speaker1= Tony Stöcker
| Speaker1_Confirmed= No
| Speaker2= Or Perlman
| Speaker2_Confirmed= No
}}



== Session description ==

In recent years, learning-based reconstruction and post-processing have transformed clinical MRI—making scans faster, sharper, and more informative. However, as data processing continues to improve, the focus is shifting back to data acquisition. With smaller artifacts becoming more apparent, we need to fix them again in acquisition. With reconstruction methods extracting all available information, the frontier of generation of new contrast features gets a revival.

This session addresses all kind of automized sequence discovery or optimization approaches - from historic EPG-based flip angle train optimization, to pulse design, and MRF schedules - trying to put it under the name "Self-learning MR". Yet, self-learning MR has many names: Optimal Control, End-to-End Optimization, Known-Operator Learning, Physics-Informed AI, Strategy Discovery, Optimal Design of Experiments, Dynamic Inverse Problems

Potential topics could be:
Self-learning MRF / CEST MRF / MRS
Self-learning trajectories
Self-learning contrast, sharpness
Self-learning over a real scanner
Self-learning imperfection mitigation
Self-learning pulses, pTX/SMS pulses
Self-learning flip angle trains, phases

While learning-based methods are well-known on the reconstruction side, we explicitly focus here on the MR preparation/acquisition side in the broadest sense. Of course, some tailored acquisitions can go hand in hand with tailored reconstruction.

Speakers:
Tony Stöcker - classic EPG-based optimizations of TSE, SFFP. pTX pulse design.
Or Perlman - Overview on MRF, autoseq, and autoCEST.
Moritz Zaiss – End-to-end sequence design

We already have an interest group mailing list with 19 groups/people, who I could invite to submit an abstract if this session gets chosen.

The program will be the following:

Option 1: (1:30h)

*Presentation by Tony Stöcker on history of EPG-based optimization + advances + pTx ''(20 min)''
* Presentation by Moritz Zaiss on end to end learing ''(10 min)''
* Presentation by Or Perlmann on AutoCEST''(10 min)''
* Selected ESMRMB Abstracts that fit to the topic: ''(50 min)''

Option 2: (1h)

* Presentation by Tony Stöcker on history of EPG-baased optimization + advances ''(15 min)''
* Presentation by N. Boulant or Tony Stöcker on Ptx pulses ''(10 min)''
* Presentation by Moritz Zaiss on end to end learing ''(15 min)''
* Presentation by Or Perlmann on AutoCEST''(15 min)''
* Discussion ''(5 min)''

Self-learning MR

2025-02-21T17:08:25Z

MoritzZaiss: /* Session description */ updated

{{Member Session
| title= Self-learning MR
| short_text= Sequence Discovery & Optimization - where we come from, where we go.
| duration= 60m
| audience= Physicists, Radiologists, AI-folks
| Moderators= Moritz Zaiss, Shaihan Malik
| Moderators_Confirmed= Zaiss, Malik
| Speaker1= Tony Stöcker
| Speaker1_Confirmed= No
| Speaker2= Or Perlman
| Speaker2_Confirmed= No
}}



== Session description ==

In recent years, learning-based reconstruction and post-processing have transformed clinical MRI—making scans faster, sharper, and more informative. However, as data processing continues to improve, the focus is shifting back to data acquisition. With smaller artifacts becoming more apparent, we need to fix them again in acquisition. With reconstruction methods extracting all available information, the frontier of generation of new contrast features gets a revival.

This session addresses all kind of automized sequence discovery or optimization approaches - from historic EPG-based flip angle train optimization, to pulse design, and MRF schedules - trying to put it under the name "Self-learning MR". Yet, self-learning MR has many names: Optimal Control, End-to-End Optimization, Known-Operator Learning, Physics-Informed AI, Strategy Discovery, Optimal Design of Experiments, Dynamic Inverse Problems

Potential topics could be:
Self-learning MRF / CEST MRF / MRS
Self-learning trajectories
Self-learning contrast, sharpness
Self-learning over a real scanner
Self-learning imperfection mitigation
Self-learning pulses, pTX/SMS pulses
Self-learning flip angle trains, phases

While learning-based methods are well-known on the reconstruction side, we explicitly focus here on the MR preparation/acquisition side in the broadest sense. Of course, some tailored acquisitions can go hand in hand with tailored reconstruction.

Speakers:
Tony Stöcker - classic EPG-based optimizations of TSE, SFFP. pTX pulse design.
Or Perlman - Overview on MRF, autoseq, and autoCEST.
Moritz Zaiss – End-to-end sequence design

We already have an interest group mailing list with 19 groups/people, who I could invite to submit an abstract if this session gets chosen.

The program will be the following:

Option 1: (1:30h)

*Presentation by Tony Stöcker on history of EPG-based optimization + advances + pTx ''(20 min)''
* Presentation by Moritz Zaiss on end to end learing ''(10 min)''
* Presentation by Or Perlmann on AutoCEST''(10 min)''
* Selected ESMRMB Abstracts that fit to the topic: ''(50 min)''

Option 2: (1h)

* Presentation by Tony Stöcker on history of EPG-baased optimization + advances ''(15 min)''
* Presentation by N. Boulant or Tony Stöcker on Ptx pulses ''(10 min)''
* Presentation by Moritz Zaiss on end to end learing ''(15 min)''
* Presentation by Or Perlmann on AutoCEST''(15 min)''
* Discussion ''(5 min)''

Self-learning MR

2025-02-21T16:58:34Z

MoritzZaiss: malik comnfirmed

{{Member Session
| title= Self-learning MR
| short_text= Sequence Discovery & Optimization - where we come from, where we go.
| duration= 60m
| audience= Physicists, Radiologists, AI-folks
| Moderators= Moritz Zaiss, Shaihan Malik
| Moderators_Confirmed= Zaiss, Malik
| Speaker1= Tony Stöcker
| Speaker1_Confirmed= No
| Speaker2= Or Perlman
| Speaker2_Confirmed= No
}}



== Session description ==

In recent years, learning-based reconstruction and post-processing have transformed clinical MRI—making scans faster, sharper, and more informative. However, as data processing continues to improve, the focus is shifting back to data acquisition. With smaller artifacts becoming more apparent, we need to fix them again in acquisition. With reconstruction methods extracting all available information, the frontier of generation of new contrast features gets a revival.

This session addresses all kind of automized sequence discovery or optimization approaches - from historic EPG-based flip angle train optimization, to pulse design, and MRF schedules - trying to put it under the name "Self-learning MR". Yet, self-learning MR has many names: Optimal Control, End-to-End Optimization, Known-Operator Learning, Physics-Informed AI, Strategy Discovery, Optimal Design of Experiments, Dynamic Inverse Problems

Potential topics could be:
Self-learning MRF / CEST MRF / MRS
Self-learning trajectories
Self-learning contrast, sharpness
Self-learning over a real scanner
Self-learning imperfection mitigation
Self-learning pulses, pTX/SMS pulses
Self-learning flip angle trains, phases

While learning-based methods are well-known on the reconstruction side, we explicitly focus here on the MR preparation/acquisition side in the broadest sense. Of course, some tailored acquisitions can go hand in hand with tailored reconstruction.

Speakers:
Tony Stöcker - classic EPG-based optimizations of TSE, SFFP. pTX pulse design.
Or Perlman - Overview on MRF, autoseq, and autoCEST.
Moritz Zaiss – End-to-end sequence design

We already have an interest group mailing list with 19 groups/people, who I could invite to submit an abstract if this session gets chosen.

The program will be the following:

* Presentation by Tony Stöcker on history of EPG-baased optimization + advances ''(15 min)''
* Presentation by N. Boulant or Tony Stöcker on Ptx pulses ''(15 min)''
* Presentation by Moritz Zaiss on end to end learing ''(15 min)''
* Presentation by Or Perlmann on AutoCEST''(15 min)''
* Discussion ''(15 min)''

Self-learning MR

2025-02-21T16:55:09Z

MoritzZaiss: This session addresses all kind of automized sequence discovery or optimization approaches - from historic EPG-based flip angle train optimization, to pulse design, and MRF schedules - trying to put it under the name "Self-learning MR".

{{Member Session
| title= Self-learning MR
| short_text= Sequence Discovery & Optimization - where we come from, where we go.
| duration= 60m
| audience= Physicists, Radiologists, AI-folks
| Moderators= Moritz Zaiss, Shaihan Malik
| Moderators_Confirmed= Zaiss
| Speaker1= Tony Stöcker
| Speaker1_Confirmed= No
| Speaker2= Or Perlman
| Speaker2_Confirmed= No
}}



== Session description ==

In recent years, learning-based reconstruction and post-processing have transformed clinical MRI—making scans faster, sharper, and more informative. However, as data processing continues to improve, the focus is shifting back to data acquisition. With smaller artifacts becoming more apparent, we need to fix them again in acquisition. With reconstruction methods extracting all available information, the frontier of generation of new contrast features gets a revival.

This session addresses all kind of automized sequence discovery or optimization approaches - from historic EPG-based flip angle train optimization, to pulse design, and MRF schedules - trying to put it under the name "Self-learning MR". Yet, self-learning MR has many names: Optimal Control, End-to-End Optimization, Known-Operator Learning, Physics-Informed AI, Strategy Discovery, Optimal Design of Experiments, Dynamic Inverse Problems

Potential topics could be:
Self-learning MRF / CEST MRF / MRS
Self-learning trajectories
Self-learning contrast, sharpness
Self-learning over a real scanner
Self-learning imperfection mitigation
Self-learning pulses, pTX/SMS pulses
Self-learning flip angle trains, phases

While learning-based methods are well-known on the reconstruction side, we explicitly focus here on the MR preparation/acquisition side in the broadest sense. Of course, some tailored acquisitions can go hand in hand with tailored reconstruction.

Speakers:
Tony Stöcker - classic EPG-based optimizations of TSE, SFFP. pTX pulse design.
Or Perlman - Overview on MRF, autoseq, and autoCEST.
Moritz Zaiss – End-to-end sequence design

We already have an interest group mailing list with 19 groups/people, who I could invite to submit an abstract if this session gets chosen.

The program will be the following:

* Presentation by Tony Stöcker on history of EPG-baased optimization + advances ''(15 min)''
* Presentation by N. Boulant or Tony Stöcker on Ptx pulses ''(15 min)''
* Presentation by Moritz Zaiss on end to end learing ''(15 min)''
* Presentation by Or Perlmann on AutoCEST''(15 min)''
* Discussion ''(15 min)''

Data crime detectives

2024-12-16T09:20:45Z

MoritzZaiss: /* Session description */

{{Member Session
| short_text= A session about data crimes in MR recon
| duration= 60m
| audience= Physicists, Radiologists
| Moderators= Moritz Zaiss, Martin Uecker
| Moderators_Confirmed= MZ confirmed
| Speaker1= Reinhard Heckel (15 min) (
| Speaker1_Confirmed= No
| Speaker2= Efrat Shimron (15 min)
|Speaker2_Confirmed=No|Speaker3=Moritz Zaiss|Speaker3_confirmed=yes}}



== Session description ==

This session will be about Data Crimes in MRI reconstructon and how simmulation can help.

Retrospective subsampling, commonly used in developing reconstruction algorithms, introduces challenges that can bias results and hinder reproducibility. When k-space data synthesized from reconstructed images, such as those from public repositories, are subsampled, hidden preprocessing pipelines (e.g., lossy compression) often enhance data sparsity. This artificially boosts algorithm performance during evaluation, leading to overly optimistic results that fail to generalize to raw data (https://doi.org/10.1073/pnas.2117203119). Another challenge emerges in transient sequences, such as MP-RAGE or FLAIR, where signal evolution depends on the history of k-space acquisitions itself. As demonstrated in recent studies (https://doi.org/10.1007/978-3-031-16446-0_67, https://doi.org/10.1007/978-3-031-72104-5_23), retrospective subsampling of these sequences fails to account for altered transient magnetization dynamics. For example, subsampling k-space lines effectively removes excitation or refocusing pulses from the prospective sequence, leading to k-space data corresponding to sharper reconstructed images than those of the original sequence. If not performed with care, this approach can cause data domain shifts or data distribution shifts (https://arxiv.org/pdf/2204.07204) between retrospectively subsampled and prospectively acquired data. To raise awareness of such "data crimes" we want to dedicate this session playing detective in public datasets, papers and hands-on simulations.

The program will be the following:

* Introduction ''(10 min)''
* Presentation by Reinhard Heckel ''(15 min) (Confirmed:No)''
* Presentation by Efrat Shimron ''(15 min) (Confirmed: Virtual Yes)''
* Presentation by Moritz Zaiss on Simulations ''(15 min) (Confirmed:Yes)''
* Questions and Answers ''(15 min)''

Data crime detectives

2024-12-16T09:17:00Z

MoritzZaiss: description

{{Member Session
| title= Example session
| short_text= A session about the future of MR
| duration= 60m
| audience= Physicists, Radiologists
| Moderators= Herman Carr, Edward Purcell
| Moderators_Confirmed= Carr only
| Speaker1= Paul Lauterbur
| Speaker1_Confirmed= Yes
| Speaker2= Peter Mansfield
}}



== Session description ==

This session will be about Data Crimes in MRI reconstructon and how simmulation can help.

Retrospective subsampling, commonly used in developing reconstruction algorithms, introduces challenges that can bias results and hinder reproducibility. When k-space data synthesized from reconstructed images, such as those from public repositories, are subsampled, hidden preprocessing pipelines (e.g., lossy compression) often enhance data sparsity. This artificially boosts algorithm performance during evaluation, leading to overly optimistic results that fail to generalize to raw data (https://doi.org/10.1073/pnas.2117203119). Another challenge emerges in transient sequences, such as MP-RAGE or FLAIR, where signal evolution depends on the history of k-space acquisitions itself. As demonstrated in recent studies (https://doi.org/10.1007/978-3-031-16446-0_67, https://doi.org/10.1007/978-3-031-72104-5_23), retrospective subsampling of these sequences fails to account for altered transient magnetization dynamics. For example, subsampling k-space lines effectively removes excitation or refocusing pulses from the prospective sequence, leading to k-space data corresponding to sharper reconstructed images than those of the original sequence. If not performed with care, this approach can cause data domain shifts or data distribution shifts (https://arxiv.org/pdf/2204.07204) between retrospectively subsampled and prospectively acquired data. To raise awareness of such "data crimes" we want to dedicate this session playing detective in public datasets, papers and hands-on simulations.

The program will be the following:

* Introduction ''(10 min)''
* Presentation by Reinhard Heckel ''(15 min) (Confirmed:No)''
* Presentation by Efrat Shimron ''(15 min) (Confirmed: Virtual Yes)''
* Presentation by Moritz Zaiss on Simulations ''(15 min) (Confirmed:Yes)''
* Questions and Answers ''(15 min)''