Results of the 2023 ISBI challenge to reduce GABA-edited MRS acquisition time. Berto, R. P., Bugler, H., Dias, G., Oliveira, M., Ueda, L., Dertkigil, S., Costa, P. D. P., Rittner, L., Merkofer, J. P., van de Sande, D. M. J., Amirrajab, S., Drenthen, G. S., Veta, M., Jansen, J. F. A., Breeuwer, M., van Sloun, R. J. G., Qayyum, A., Rodero, C., Niederer, S., Souza, R., & Harris, A. D. MAGMA, 2024. Berto, Rodrigo Pommot Bugler, Hanna Dias, Gabriel Oliveira, Mateus Ueda, Lucas Dertkigil, Sergio Costa, Paula D P Rittner, Leticia Merkofer, Julian P van de Sande, Dennis M J Amirrajab, Sina Drenthen, Gerhard S Veta, Mitko Jansen, Jacobus F A Breeuwer, Marcel van Sloun, Ruud J G Qayyum, Abdul Rodero, Cristobal Niederer, Steven Souza, Roberto Harris, Ashley D eng RG/20/4/34 803/BHF_/British Heart Foundation/United Kingdom Germany 2024/04/13 20:42 MAGMA. 2024 Apr 13. doi: 10.1007/s10334-024-01156-9.![Results of the 2023 ISBI challenge to reduce GABA-edited MRS acquisition time [link]](https://bibbase.org/img/filetypes/link.svg "link")
Paper doi abstract bibtex PURPOSE: Use a conference challenge format to compare machine learning-based gamma-aminobutyric acid (GABA)-edited magnetic resonance spectroscopy (MRS) reconstruction models using one-quarter of the transients typically acquired during a complete scan. METHODS: There were three tracks: Track 1: simulated data, Track 2: identical acquisition parameters with in vivo data, and Track 3: different acquisition parameters with in vivo data. The mean squared error, signal-to-noise ratio, linewidth, and a proposed shape score metric were used to quantify model performance. Challenge organizers provided open access to a baseline model, simulated noise-free data, guides for adding synthetic noise, and in vivo data. RESULTS: Three submissions were compared. A covariance matrix convolutional neural network model was most successful for Track 1. A vision transformer model operating on a spectrogram data representation was most successful for Tracks 2 and 3. Deep learning (DL) reconstructions with 80 transients achieved equivalent or better SNR, linewidth and fit error compared to conventional 320 transient reconstructions. However, some DL models optimized linewidth and SNR without actually improving overall spectral quality, indicating a need for more robust metrics. CONCLUSION: DL-based reconstruction pipelines have the promise to reduce the number of transients required for GABA-edited MRS.
@article{RN356,
author = {Berto, R. P. and Bugler, H. and Dias, G. and Oliveira, M. and Ueda, L. and Dertkigil, S. and Costa, P. D. P. and Rittner, L. and Merkofer, J. P. and van de Sande, D. M. J. and Amirrajab, S. and Drenthen, G. S. and Veta, M. and Jansen, J. F. A. and Breeuwer, M. and van Sloun, R. J. G. and Qayyum, A. and Rodero, C. and Niederer, S. and Souza, R. and Harris, A. D.},
title = {Results of the 2023 ISBI challenge to reduce GABA-edited MRS acquisition time},
journal = {MAGMA},
note = {Berto, Rodrigo Pommot
Bugler, Hanna
Dias, Gabriel
Oliveira, Mateus
Ueda, Lucas
Dertkigil, Sergio
Costa, Paula D P
Rittner, Leticia
Merkofer, Julian P
van de Sande, Dennis M J
Amirrajab, Sina
Drenthen, Gerhard S
Veta, Mitko
Jansen, Jacobus F A
Breeuwer, Marcel
van Sloun, Ruud J G
Qayyum, Abdul
Rodero, Cristobal
Niederer, Steven
Souza, Roberto
Harris, Ashley D
eng
RG/20/4/34 803/BHF_/British Heart Foundation/United Kingdom
Germany
2024/04/13 20:42
MAGMA. 2024 Apr 13. doi: 10.1007/s10334-024-01156-9.},
abstract = {PURPOSE: Use a conference challenge format to compare machine learning-based gamma-aminobutyric acid (GABA)-edited magnetic resonance spectroscopy (MRS) reconstruction models using one-quarter of the transients typically acquired during a complete scan. METHODS: There were three tracks: Track 1: simulated data, Track 2: identical acquisition parameters with in vivo data, and Track 3: different acquisition parameters with in vivo data. The mean squared error, signal-to-noise ratio, linewidth, and a proposed shape score metric were used to quantify model performance. Challenge organizers provided open access to a baseline model, simulated noise-free data, guides for adding synthetic noise, and in vivo data. RESULTS: Three submissions were compared. A covariance matrix convolutional neural network model was most successful for Track 1. A vision transformer model operating on a spectrogram data representation was most successful for Tracks 2 and 3. Deep learning (DL) reconstructions with 80 transients achieved equivalent or better SNR, linewidth and fit error compared to conventional 320 transient reconstructions. However, some DL models optimized linewidth and SNR without actually improving overall spectral quality, indicating a need for more robust metrics. CONCLUSION: DL-based reconstruction pipelines have the promise to reduce the number of transients required for GABA-edited MRS.},
keywords = {Benchmarking
Computer
Deep learning
Magnetic resonance spectroscopy
Neural networks},
ISSN = {1352-8661 (Electronic)
0968-5243 (Linking)},
DOI = {10.1007/s10334-024-01156-9},
url = {https://www.ncbi.nlm.nih.gov/pubmed/38613715},
year = {2024},
type = {Journal Article}
}
Downloads: 0
{"_id":"QmJdZwzJe7EzaqaHW","bibbaseid":"berto-bugler-dias-oliveira-ueda-dertkigil-costa-rittner-etal-resultsofthe2023isbichallengetoreducegabaeditedmrsacquisitiontime-2024","author_short":["Berto, R. P.","Bugler, H.","Dias, G.","Oliveira, M.","Ueda, L.","Dertkigil, S.","Costa, P. D. P.","Rittner, L.","Merkofer, J. P.","van de Sande, D. M. J.","Amirrajab, S.","Drenthen, G. S.","Veta, M.","Jansen, J. F. A.","Breeuwer, M.","van Sloun, R. J. G.","Qayyum, A.","Rodero, C.","Niederer, S.","Souza, R.","Harris, A. D."],"bibdata":{"bibtype":"article","type":"Journal Article","author":[{"propositions":[],"lastnames":["Berto"],"firstnames":["R.","P."],"suffixes":[]},{"propositions":[],"lastnames":["Bugler"],"firstnames":["H."],"suffixes":[]},{"propositions":[],"lastnames":["Dias"],"firstnames":["G."],"suffixes":[]},{"propositions":[],"lastnames":["Oliveira"],"firstnames":["M."],"suffixes":[]},{"propositions":[],"lastnames":["Ueda"],"firstnames":["L."],"suffixes":[]},{"propositions":[],"lastnames":["Dertkigil"],"firstnames":["S."],"suffixes":[]},{"propositions":[],"lastnames":["Costa"],"firstnames":["P.","D.","P."],"suffixes":[]},{"propositions":[],"lastnames":["Rittner"],"firstnames":["L."],"suffixes":[]},{"propositions":[],"lastnames":["Merkofer"],"firstnames":["J.","P."],"suffixes":[]},{"propositions":["van","de"],"lastnames":["Sande"],"firstnames":["D.","M.","J."],"suffixes":[]},{"propositions":[],"lastnames":["Amirrajab"],"firstnames":["S."],"suffixes":[]},{"propositions":[],"lastnames":["Drenthen"],"firstnames":["G.","S."],"suffixes":[]},{"propositions":[],"lastnames":["Veta"],"firstnames":["M."],"suffixes":[]},{"propositions":[],"lastnames":["Jansen"],"firstnames":["J.","F.","A."],"suffixes":[]},{"propositions":[],"lastnames":["Breeuwer"],"firstnames":["M."],"suffixes":[]},{"propositions":["van"],"lastnames":["Sloun"],"firstnames":["R.","J.","G."],"suffixes":[]},{"propositions":[],"lastnames":["Qayyum"],"firstnames":["A."],"suffixes":[]},{"propositions":[],"lastnames":["Rodero"],"firstnames":["C."],"suffixes":[]},{"propositions":[],"lastnames":["Niederer"],"firstnames":["S."],"suffixes":[]},{"propositions":[],"lastnames":["Souza"],"firstnames":["R."],"suffixes":[]},{"propositions":[],"lastnames":["Harris"],"firstnames":["A.","D."],"suffixes":[]}],"title":"Results of the 2023 ISBI challenge to reduce GABA-edited MRS acquisition time","journal":"MAGMA","note":"Berto, Rodrigo Pommot Bugler, Hanna Dias, Gabriel Oliveira, Mateus Ueda, Lucas Dertkigil, Sergio Costa, Paula D P Rittner, Leticia Merkofer, Julian P van de Sande, Dennis M J Amirrajab, Sina Drenthen, Gerhard S Veta, Mitko Jansen, Jacobus F A Breeuwer, Marcel van Sloun, Ruud J G Qayyum, Abdul Rodero, Cristobal Niederer, Steven Souza, Roberto Harris, Ashley D eng RG/20/4/34 803/BHF_/British Heart Foundation/United Kingdom Germany 2024/04/13 20:42 MAGMA. 2024 Apr 13. doi: 10.1007/s10334-024-01156-9.","abstract":"PURPOSE: Use a conference challenge format to compare machine learning-based gamma-aminobutyric acid (GABA)-edited magnetic resonance spectroscopy (MRS) reconstruction models using one-quarter of the transients typically acquired during a complete scan. METHODS: There were three tracks: Track 1: simulated data, Track 2: identical acquisition parameters with in vivo data, and Track 3: different acquisition parameters with in vivo data. The mean squared error, signal-to-noise ratio, linewidth, and a proposed shape score metric were used to quantify model performance. Challenge organizers provided open access to a baseline model, simulated noise-free data, guides for adding synthetic noise, and in vivo data. RESULTS: Three submissions were compared. A covariance matrix convolutional neural network model was most successful for Track 1. A vision transformer model operating on a spectrogram data representation was most successful for Tracks 2 and 3. Deep learning (DL) reconstructions with 80 transients achieved equivalent or better SNR, linewidth and fit error compared to conventional 320 transient reconstructions. However, some DL models optimized linewidth and SNR without actually improving overall spectral quality, indicating a need for more robust metrics. CONCLUSION: DL-based reconstruction pipelines have the promise to reduce the number of transients required for GABA-edited MRS.","keywords":"Benchmarking Computer Deep learning Magnetic resonance spectroscopy Neural networks","issn":"1352-8661 (Electronic) 0968-5243 (Linking)","doi":"10.1007/s10334-024-01156-9","url":"https://www.ncbi.nlm.nih.gov/pubmed/38613715","year":"2024","bibtex":"@article{RN356,\n author = {Berto, R. P. and Bugler, H. and Dias, G. and Oliveira, M. and Ueda, L. and Dertkigil, S. and Costa, P. D. P. and Rittner, L. and Merkofer, J. P. and van de Sande, D. M. J. and Amirrajab, S. and Drenthen, G. S. and Veta, M. and Jansen, J. F. A. and Breeuwer, M. and van Sloun, R. J. G. and Qayyum, A. and Rodero, C. and Niederer, S. and Souza, R. and Harris, A. D.},\n title = {Results of the 2023 ISBI challenge to reduce GABA-edited MRS acquisition time},\n journal = {MAGMA},\n note = {Berto, Rodrigo Pommot\nBugler, Hanna\nDias, Gabriel\nOliveira, Mateus\nUeda, Lucas\nDertkigil, Sergio\nCosta, Paula D P\nRittner, Leticia\nMerkofer, Julian P\nvan de Sande, Dennis M J\nAmirrajab, Sina\nDrenthen, Gerhard S\nVeta, Mitko\nJansen, Jacobus F A\nBreeuwer, Marcel\nvan Sloun, Ruud J G\nQayyum, Abdul\nRodero, Cristobal\nNiederer, Steven\nSouza, Roberto\nHarris, Ashley D\neng\nRG/20/4/34 803/BHF_/British Heart Foundation/United Kingdom\nGermany\n2024/04/13 20:42\nMAGMA. 2024 Apr 13. doi: 10.1007/s10334-024-01156-9.},\n abstract = {PURPOSE: Use a conference challenge format to compare machine learning-based gamma-aminobutyric acid (GABA)-edited magnetic resonance spectroscopy (MRS) reconstruction models using one-quarter of the transients typically acquired during a complete scan. METHODS: There were three tracks: Track 1: simulated data, Track 2: identical acquisition parameters with in vivo data, and Track 3: different acquisition parameters with in vivo data. The mean squared error, signal-to-noise ratio, linewidth, and a proposed shape score metric were used to quantify model performance. Challenge organizers provided open access to a baseline model, simulated noise-free data, guides for adding synthetic noise, and in vivo data. RESULTS: Three submissions were compared. A covariance matrix convolutional neural network model was most successful for Track 1. A vision transformer model operating on a spectrogram data representation was most successful for Tracks 2 and 3. Deep learning (DL) reconstructions with 80 transients achieved equivalent or better SNR, linewidth and fit error compared to conventional 320 transient reconstructions. However, some DL models optimized linewidth and SNR without actually improving overall spectral quality, indicating a need for more robust metrics. CONCLUSION: DL-based reconstruction pipelines have the promise to reduce the number of transients required for GABA-edited MRS.},\n keywords = {Benchmarking\nComputer\nDeep learning\nMagnetic resonance spectroscopy\nNeural networks},\n ISSN = {1352-8661 (Electronic)\n0968-5243 (Linking)},\n DOI = {10.1007/s10334-024-01156-9},\n url = {https://www.ncbi.nlm.nih.gov/pubmed/38613715},\n year = {2024},\n type = {Journal Article}\n}\n\n","author_short":["Berto, R. P.","Bugler, H.","Dias, G.","Oliveira, M.","Ueda, L.","Dertkigil, S.","Costa, P. D. P.","Rittner, L.","Merkofer, J. P.","van de Sande, D. M. J.","Amirrajab, S.","Drenthen, G. S.","Veta, M.","Jansen, J. F. A.","Breeuwer, M.","van Sloun, R. J. G.","Qayyum, A.","Rodero, C.","Niederer, S.","Souza, R.","Harris, A. D."],"key":"RN356","id":"RN356","bibbaseid":"berto-bugler-dias-oliveira-ueda-dertkigil-costa-rittner-etal-resultsofthe2023isbichallengetoreducegabaeditedmrsacquisitiontime-2024","role":"author","urls":{"Paper":"https://www.ncbi.nlm.nih.gov/pubmed/38613715"},"keyword":["Benchmarking Computer Deep learning Magnetic resonance spectroscopy Neural networks"],"metadata":{"authorlinks":{}},"html":""},"bibtype":"article","biburl":"https://raw.githubusercontent.com/jansenjfa1/bibbase.github.io/master/jansenjfa.bib","dataSources":["TCkfRWJAZvbLAZi29"],"keywords":["benchmarking computer deep learning magnetic resonance spectroscopy neural networks"],"search_terms":["results","2023","isbi","challenge","reduce","gaba","edited","mrs","acquisition","time","berto","bugler","dias","oliveira","ueda","dertkigil","costa","rittner","merkofer","van de sande","amirrajab","drenthen","veta","jansen","breeuwer","van sloun","qayyum","rodero","niederer","souza","harris"],"title":"Results of the 2023 ISBI challenge to reduce GABA-edited MRS acquisition time","year":2024}