Pub Date : 2024-07-05DOI: 10.1101/2024.07.03.24309779
Su Hwan Kim, Severin Schramm, Jonas Wihl, Philipp Raffler, Marlene Tahedl, Julian Canisius, Ina Luiken, Lukas Endroes, Stefan Reischl, Alexander Marka, Robert Walter, Mathias Schillmaier, Claus Zimmer, Benedikt Wiestler, Dennis Martin Hedderich
Pub Date : 2024-07-04DOI: 10.1101/2024.07.03.24309882
Vadym Gryshchuk, Devesh Singh, Stefan J. Teipel, Martin Dyrba
Neurodegenerative diseases such as Alzheimer's disease (AD) or frontotemporal lobar degeneration (FTLD) involve specific loss of brain volume, detectable in vivo using T1-weighted MRI scans. Supervised machine learning approaches classifying neurodegenerative diseases require diagnostic-labels for each sample. However, it can be difficult to obtain expert labels for a large amount of data. Self-supervised learning (SSL) offers an alternative for training machine learning models without data-labels. We investigated if the SSL models can applied to distinguish between different neurodegenerative disorders in an interpretable manner. Our method comprises a feature extractor and a downstream classification head. A deep convolutional neural network trained in a contrastive self-supervised way serves as the feature extractor, learning latent representation, while the classifier head is a single-layer perceptron. We used N=2694 T1-weighted MRI scans from four data cohorts: two ADNI datasets, AIBL and FTLDNI, including cognitively normal controls (CN), cases with prodromal and clinical AD, as well as FTLD cases differentiated into its sub-types. Our results showed that the feature extractor trained in a self-supervised way provides generalizable and robust representations for the downstream classification. For AD vs. CN, our model achieves 82% balanced accuracy on the test subset and 80% on an independent holdout dataset. Similarly, the behavioral variant of frontotemporal dementia (BV) vs. CN model attains an 88% balanced accuracy on the test subset. The average feature attribution heatmaps obtained by the Integrated Gradient method highlighted hallmark regions, i.e., temporal gray matter atrophy for AD, and insular atrophy for BV. In conclusion, our models perform comparably to state-of-the-art supervised deep learning approaches. This suggests that the SSL methodology can successfully make use of unannotated neuroimaging datasets as training data while remaining robust and interpretable.
{"title":"Contrastive Self-supervised Learning for Neurodegenerative Disorder Classification","authors":"Vadym Gryshchuk, Devesh Singh, Stefan J. Teipel, Martin Dyrba","doi":"10.1101/2024.07.03.24309882","DOIUrl":"https://doi.org/10.1101/2024.07.03.24309882","url":null,"abstract":"Neurodegenerative diseases such as Alzheimer's disease (AD) or frontotemporal lobar degeneration (FTLD) involve specific loss of brain volume, detectable in vivo using T1-weighted MRI scans. Supervised machine learning approaches classifying neurodegenerative diseases require diagnostic-labels for each sample. However, it can be difficult to obtain expert labels for a large amount of data. Self-supervised learning (SSL) offers an alternative for training machine learning models without data-labels. We investigated if the SSL models can applied to distinguish between different neurodegenerative disorders in an interpretable manner. Our method comprises a feature extractor and a downstream classification head. A deep convolutional neural network trained in a contrastive self-supervised way serves as the feature extractor, learning latent representation, while the classifier head is a single-layer perceptron. We used N=2694 T1-weighted MRI scans from four data cohorts: two ADNI datasets, AIBL and FTLDNI, including cognitively normal controls (CN), cases with prodromal and clinical AD, as well as FTLD cases differentiated into its sub-types. Our results showed that the feature extractor trained in a self-supervised way provides generalizable and robust representations for the downstream classification. For AD vs. CN, our model achieves 82% balanced accuracy on the test subset and 80% on an independent holdout dataset. Similarly, the behavioral variant of frontotemporal dementia (BV) vs. CN model attains an 88% balanced accuracy on the test subset. The average feature attribution heatmaps obtained by the Integrated Gradient method highlighted hallmark regions, i.e., temporal gray matter atrophy for AD, and insular atrophy for BV. In conclusion, our models perform comparably to state-of-the-art supervised deep learning approaches. This suggests that the SSL methodology can successfully make use of unannotated neuroimaging datasets as training data while remaining robust and interpretable.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"13 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-07-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141550997","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-07-03DOI: 10.1101/2024.07.02.24309338
Min Wang, Tao Chen, Zhongyi He, Lawrence Wing-Chi Chan, qinger guo, Shuyang Cai, Jingfeng Duan, Danbin Zhang, Xunda Wang, Yu Fang, Hong Yang
Major depressive disorder (MDD) is characterized by disrupted functional network connectivity (FNC), with unclear underlying dynamics. We investigated both static FNC (sFNC) and dynamic FNC (dFNC) on resting-state fMRI data from drug-naive first-episode MDD patients and healthy controls (HC). MDD patients exhibited lower sFNC within and between sensory and motor networks than HC. Four dFNC states were identified, including a globally-weakly-connected state, a cognitive-control-dominated state, a globally-positively-connected state, and an antagonistic state. The antagonistic state was marked by strong positive connections within the sensorimotor domain and their anti-correlations with the executive-motor control domain. Notably, MDD patients exhibited significantly longer time dwelling in the globally-weakly-connected state, at the cost of significantly shorter time dwelling in the antagonistic state. Further, only the mean dwell time of this antagonistic state was significantly anticorrelated to disease severity measures. Our study highlights the altered dynamics of the antagonistic state as a fundamental aspect of disrupted FNC in early MDD.
{"title":"Altered dynamic functional connectivity in antagonistic state in first-episode, drug-naive patients with major depressive disorder.","authors":"Min Wang, Tao Chen, Zhongyi He, Lawrence Wing-Chi Chan, qinger guo, Shuyang Cai, Jingfeng Duan, Danbin Zhang, Xunda Wang, Yu Fang, Hong Yang","doi":"10.1101/2024.07.02.24309338","DOIUrl":"https://doi.org/10.1101/2024.07.02.24309338","url":null,"abstract":"Major depressive disorder (MDD) is characterized by disrupted functional network connectivity (FNC), with unclear underlying dynamics. We investigated both static FNC (sFNC) and dynamic FNC (dFNC) on resting-state fMRI data from drug-naive first-episode MDD patients and healthy controls (HC). MDD patients exhibited lower sFNC within and between sensory and motor networks than HC. Four dFNC states were identified, including a globally-weakly-connected state, a cognitive-control-dominated state, a globally-positively-connected state, and an antagonistic state. The antagonistic state was marked by strong positive connections within the sensorimotor domain and their anti-correlations with the executive-motor control domain. Notably, MDD patients exhibited significantly longer time dwelling in the globally-weakly-connected state, at the cost of significantly shorter time dwelling in the antagonistic state. Further, only the mean dwell time of this antagonistic state was significantly anticorrelated to disease severity measures. Our study highlights the altered dynamics of the antagonistic state as a fundamental aspect of disrupted FNC in early MDD.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"39 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141550998","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-07-01DOI: 10.1101/2024.06.28.24307535
Ayşe Sıla Dokumacı, Katy Vecchiato, Raphael Tomi-Tricot, Michael Eyre, Philippa Bridgen, Pierluigi Di Cio, Chiara Casella, Tobias C. Wood, Jan Sedlacik, Tom Wilkinson, Sharon L. Giles, Joseph V. Hajnal, Jonathan O'Muircheartaigh, Shaihan J. Malik, David W. Carmichael
Introduction Quantitative MRI is important for non-invasive tissue characterisation. In previous work we developed a clinically feasible multi-contrast protocol for T1-weighted imaging based on the MP2RAGE sequence that was optimised for both children and adults. It was demonstrated that a range of Fluid And White Matter Suppression (FLAWS) related contrasts could be produced while maintaining T1-weighted uniform image (UNI) quality, a challenge at higher field strengths. Here we introduce an approach to use these images to calculate effective proton density (PD*) and quantitative T1 relaxation maps especially for shorter repetition times (TRMP2RAGE) than those typically used previously. Methods T1 and PD* were estimated from the analytical equations of the MP2RAGE signal derived for partial Fourier acquisitions. The sensitivity of the fitting results was evaluated with respect to the TRMP2RAGE and B1+ effects on both excitation flip angles and inversion efficiency and compared to vendor T1 maps which do not use B1+ information. Data acquired for a range of individuals (aged 10-54 years) at the shortest TRMP2RAGE (4000ms) were compared across white matter (WM), cortical grey matter, and deep grey matter regions. Results The T1 values were insensitive to the choice of different TRMP2RAGE. The results were similar to the vendor T1 maps if the B1+ effects on the excitation flip angle and inversion efficiency were not included in the fits. T1 values varied over development into adulthood, especially for the deep grey matter regions whereas only a very small difference was observed for WM T1. Effective PD maps were produced which did not show a significant difference between children and adults for the age range included. Conclusion We produced PD* maps and improved the accuracy of T1 maps from an MP2RAGE protocol that is optimised for UNI and FLAWS-related contrasts in a single scan at 7T by incorporating the excitation flip angle and inversion efficiency related effects of B1+ in the fitting. This multi-parametric protocol made it possible to acquire high resolution images (0.65mm iso) in children and adults within a clinically feasible duration (7:18 min:s). The combination of analytical equations utilizing B1+ maps led to T1 fits that were consistent at different TRMP2RAGE values. Average WM T1 values of adults and children were very similar (1092ms vs 1117ms) while expected reductions in T1 with age were found for GM especially for deep GM.
{"title":"Quantitative T1 and Effective Proton Density (PD*) mapping in children and adults at 7T from an MP2RAGE sequence optimised for uniform T1-weighted (UNI) and FLuid And White matter Suppression (FLAWS) contrasts","authors":"Ayşe Sıla Dokumacı, Katy Vecchiato, Raphael Tomi-Tricot, Michael Eyre, Philippa Bridgen, Pierluigi Di Cio, Chiara Casella, Tobias C. Wood, Jan Sedlacik, Tom Wilkinson, Sharon L. Giles, Joseph V. Hajnal, Jonathan O'Muircheartaigh, Shaihan J. Malik, David W. Carmichael","doi":"10.1101/2024.06.28.24307535","DOIUrl":"https://doi.org/10.1101/2024.06.28.24307535","url":null,"abstract":"<strong>Introduction</strong>\u0000Quantitative MRI is important for non-invasive tissue characterisation. In previous work we developed a clinically feasible multi-contrast protocol for T<sub>1</sub>-weighted imaging based on the MP2RAGE sequence that was optimised for both children and adults. It was demonstrated that a range of Fluid And White Matter Suppression (FLAWS) related contrasts could be produced while maintaining T<sub>1</sub>-weighted uniform image (UNI) quality, a challenge at higher field strengths. Here we introduce an approach to use these images to calculate effective proton density (PD<sup>*</sup>) and quantitative T<sub>1</sub> relaxation maps especially for shorter repetition times (TR<sub>MP2RAGE</sub>) than those typically used previously.\u0000<strong>Methods</strong>\u0000T<sub>1</sub> and PD<sup>*</sup> were estimated from the analytical equations of the MP2RAGE signal derived for partial Fourier acquisitions. The sensitivity of the fitting results was evaluated with respect to the TR<sub>MP2RAGE</sub> and B<sub>1</sub><sup>+</sup> effects on both excitation flip angles and inversion efficiency and compared to vendor T<sub>1</sub> maps which do not use B<sub>1</sub><sup>+</sup> information. Data acquired for a range of individuals (aged 10-54 years) at the shortest TR<sub>MP2RAGE</sub> (4000ms) were compared across white matter (WM), cortical grey matter, and deep grey matter regions. <strong>Results</strong> The T<sub>1</sub> values were insensitive to the choice of different TR<sub>MP2RAGE</sub>. The results were similar to the vendor T<sub>1</sub> maps if the B<sub>1</sub><sup>+</sup> effects on the excitation flip angle and inversion efficiency were not included in the fits. T<sub>1</sub> values varied over development into adulthood, especially for the deep grey matter regions whereas only a very small difference was observed for WM T<sub>1</sub>. Effective PD maps were produced which did not show a significant difference between children and adults for the age range included. <strong>Conclusion</strong>\u0000We produced PD<sup>*</sup> maps and improved the accuracy of T<sub>1</sub> maps from an MP2RAGE protocol that is optimised for UNI and FLAWS-related contrasts in a single scan at 7T by incorporating the excitation flip angle and inversion efficiency related effects of B<sub>1</sub><sup>+</sup> in the fitting. This multi-parametric protocol made it possible to acquire high resolution images (0.65mm iso) in children and adults within a clinically feasible duration (7:18 min:s). The combination of analytical equations utilizing B<sub>1</sub><sup>+</sup> maps led to T<sub>1</sub> fits that were consistent at different TR<sub>MP2RAGE</sub> values. Average WM T<sub>1</sub> values of adults and children were very similar (1092ms vs 1117ms) while expected reductions in T<sub>1</sub> with age were found for GM especially for deep GM.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"22 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503248","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-28DOI: 10.1101/2024.06.27.24309589
David Izquierdo-Garcia, Pauline Désogère, Anne L. Philip, David E. Sosnovik, Ciprian Catana, Peter Caravan
Purpose: This study presents the biodistribution, clearance and dosimetry estimates of [64Cu]FBP8 Binding Probe #8 ([64Cu]FBP8) in healthy subjects. Procedures: This prospective study included 8 healthy subjects to evaluate biodistribution, safety and dosimetry estimates of [64Cu]FBP8, a fibrin-binding positron emission tomography (PET) probe. All subjects underwent up to 3 sessions of PET/Magnetic Resonance Imaging (PET/MRI) 0-2 hours, 4h and 24h post injection. Dosimetry estimates were obtained using OLINDA 2.2 software. Results: Subjects were injected with ~400 MBq of [64Cu]FBP8. Subjects did not experience adverse effects due to the injection of the probe. [64Cu]FBP8 PET images demonstrated fast blood clearance (half-life = 67 min) and renal excretion of the probe, showing low background signal across the body. The organs with the higher doses were: the urinary bladder (0.075 vs. 0.091 mGy/MBq for males and females, respectively); the kidneys (0.050 vs. 0.056 mGy/MBq respectively); and the liver (0.027 vs. 0.035 mGy/MBq respectively). The combined mean effective dose for males and females was 0.016 ± 0.0029 mSv/MBq, lower than the widely used [18F]fluorodeoxyglucose ([18F]FDG, 0.020mSv/MBq). Conclusions: This study demonstrates the following properties of the [64Cu]FBP8 probe: low dosimetry estimates; fast blood clearance and renal excretion; low background signal; and whole-body acquisition within 20 minutes in a single session. These properties provide the basis for [64Cu]FBP8 to be an excellent candidate for whole-body non-invasive imaging of fibrin, an important driver/feature in many cardiovascular, oncological and neurological conditions.
{"title":"Dosimetry of [64Cu]FBP8: a fibrin-binding PET probe","authors":"David Izquierdo-Garcia, Pauline Désogère, Anne L. Philip, David E. Sosnovik, Ciprian Catana, Peter Caravan","doi":"10.1101/2024.06.27.24309589","DOIUrl":"https://doi.org/10.1101/2024.06.27.24309589","url":null,"abstract":"Purpose: This study presents the biodistribution, clearance and dosimetry estimates of [64Cu]FBP8 Binding Probe #8 ([64Cu]FBP8) in healthy subjects.\u0000Procedures: This prospective study included 8 healthy subjects to evaluate biodistribution, safety and dosimetry estimates of [64Cu]FBP8, a fibrin-binding positron emission tomography (PET) probe. All subjects underwent up to 3 sessions of PET/Magnetic Resonance Imaging (PET/MRI) 0-2 hours, 4h and 24h post injection. Dosimetry estimates were obtained using OLINDA 2.2 software.\u0000Results: Subjects were injected with ~400 MBq of [64Cu]FBP8. Subjects did not experience adverse effects due to the injection of the probe. [64Cu]FBP8 PET images demonstrated fast blood clearance (half-life = 67 min) and renal excretion of the probe, showing low background signal across the body. The organs with the higher doses were: the urinary bladder (0.075 vs. 0.091 mGy/MBq for males and females, respectively); the kidneys (0.050 vs. 0.056 mGy/MBq respectively); and the liver (0.027 vs. 0.035 mGy/MBq respectively). The combined mean effective dose for males and females was 0.016 ± 0.0029 mSv/MBq, lower than the widely used [18F]fluorodeoxyglucose ([18F]FDG, 0.020mSv/MBq).\u0000Conclusions: This study demonstrates the following properties of the [64Cu]FBP8 probe: low dosimetry estimates; fast blood clearance and renal excretion; low background signal; and whole-body acquisition within 20 minutes in a single session. These properties provide the basis for [64Cu]FBP8 to be an excellent candidate for whole-body non-invasive imaging of fibrin, an important driver/feature in many cardiovascular, oncological and neurological conditions.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"23 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503249","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-25DOI: 10.1101/2024.06.24.24309431
Julia Kiewitz, Orhun Utku Aydin, Adam Hilbert, Marie Gultom, Anouar Nouri, Ahmed A Khalil, Peter Vajkoczy, Satoru Tanioka, Fujimaro Ishida, Nora F. Dengler, Dietmar Frey
Introduction Aneurysmal subarachnoid hemorrhage (aSAH) is a life-threatening condition with a significant variability in patients’ outcomes. Radiographic scores used to assess the extent of SAH or other potentially outcome-relevant pathologies are limited by interrater variability and do not utilize all available information from the imaging. Image segmentation plays an important role in extracting relevant information from images by enabling precise identification and delineation of objects or regions of interest. Thus, segmentation offers the potential for automatization of score assessments and downstream outcome prediction using precise volumetric information. Our study aims to develop a deep learning model that enables automated multiclass segmentation of structures and pathologies relevant for aSAH outcome prediction.
{"title":"Deep Learning-based Multiclass Segmentation in Aneurysmal Subarachnoid Hemorrhage","authors":"Julia Kiewitz, Orhun Utku Aydin, Adam Hilbert, Marie Gultom, Anouar Nouri, Ahmed A Khalil, Peter Vajkoczy, Satoru Tanioka, Fujimaro Ishida, Nora F. Dengler, Dietmar Frey","doi":"10.1101/2024.06.24.24309431","DOIUrl":"https://doi.org/10.1101/2024.06.24.24309431","url":null,"abstract":"<strong>Introduction</strong> Aneurysmal subarachnoid hemorrhage (aSAH) is a life-threatening condition with a significant variability in patients’ outcomes. Radiographic scores used to assess the extent of SAH or other potentially outcome-relevant pathologies are limited by interrater variability and do not utilize all available information from the imaging. Image segmentation plays an important role in extracting relevant information from images by enabling precise identification and delineation of objects or regions of interest. Thus, segmentation offers the potential for automatization of score assessments and downstream outcome prediction using precise volumetric information. Our study aims to develop a deep learning model that enables automated multiclass segmentation of structures and pathologies relevant for aSAH outcome prediction.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"2013 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141528745","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-25DOI: 10.1101/2024.06.25.24309404
Andrzej Liebert, Dominique Hadler, Hannes Schreiter, Chris Ehring, Luise Brock, Lorenz A. Kapsner, Jessica Eberle, Ramona Erber, Julius Emons, Frederik B. Laun, Michael Uder, Evelyn Wenkel, Sabine Ohlmeyer, Sebastian Bickelhaupt
Background: Breast magnetic resonance imaging (MRI) protocols often include T2-weighted fat-saturated (T2w-FS) sequences, which are vital for tissue characterization but significantly increase scan time. Purpose: This study aims to evaluate whether a 2D-U-Net neural network can generate virtual T2w-FS images from routine multiparametric breast MRI sequences. Materials and Methods: This IRB approved, retrospective study included n=914 breast MRI examinations performed between January 2017 and June 2020. The dataset was divided into training (n=665), validation (n=74), and test sets (n=175). The U-Net was trained on T1-weighted (T1w), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced (DCE) sequences to generate virtual T2w-FS images (VirtuT2). Quantitative metrics and a qualitative multi-reader assessment by two radiologists were used to evaluate the VirtuT2 images. Results: VirtuT2 images demonstrated high structural similarity (SSIM=0.87) and peak signal-to-noise ratio (PSNR=24.90) compared to original T2w-FS images. High level of the frequency error norm (HFNE=0.87) indicates strong blurring presence in the VirtuT2 images, which was also confirmed in qualitative reading. Radiologists correctly identified VirtuT2 images with 92.3% and 94.2% accuracy, respectively. No significant difference in diagnostic image quality (DIQ) was noted for one reader (p=0.21), while the other reported significantly lower DIQ for VirtuT2 (p<=0.001). Moderate inter-reader agreement was observed for edema detection on T2w-FS images (ƙ=0.43), decreasing to fair on VirtuT2 images (ƙ=0.36). Conclusion: The 2D-U-Net can technically generate virtual T2w-FS images with high similarity to real T2w-FS images, though blurring remains a limitation. Further investigation of other architectures and using larger datasets are needed to improve clinical applicability.
{"title":"Feasibility to virtually generate T2 fat-saturated breast MRI by convolutional neural networks","authors":"Andrzej Liebert, Dominique Hadler, Hannes Schreiter, Chris Ehring, Luise Brock, Lorenz A. Kapsner, Jessica Eberle, Ramona Erber, Julius Emons, Frederik B. Laun, Michael Uder, Evelyn Wenkel, Sabine Ohlmeyer, Sebastian Bickelhaupt","doi":"10.1101/2024.06.25.24309404","DOIUrl":"https://doi.org/10.1101/2024.06.25.24309404","url":null,"abstract":"Background: Breast magnetic resonance imaging (MRI) protocols often include T2-weighted fat-saturated (T2w-FS) sequences, which are vital for tissue characterization but significantly increase scan time. Purpose: This study aims to evaluate whether a 2D-U-Net neural network can generate virtual T2w-FS images from routine multiparametric breast MRI sequences.\u0000Materials and Methods: This IRB approved, retrospective study included n=914 breast MRI examinations performed between January 2017 and June 2020. The dataset was divided into training (n=665), validation (n=74), and test sets (n=175). The U-Net was trained on T1-weighted (T1w), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced (DCE) sequences to generate virtual T2w-FS images (VirtuT2). Quantitative metrics and a qualitative multi-reader assessment by two radiologists were used to evaluate the VirtuT2 images.\u0000Results: VirtuT2 images demonstrated high structural similarity (SSIM=0.87) and peak signal-to-noise ratio (PSNR=24.90) compared to original T2w-FS images. High level of the frequency error norm (HFNE=0.87) indicates strong blurring presence in the VirtuT2 images, which was also confirmed in qualitative reading. Radiologists correctly identified VirtuT2 images with 92.3% and 94.2% accuracy, respectively. No significant difference in diagnostic image quality (DIQ) was noted for one reader (p=0.21), while the other reported significantly lower DIQ for VirtuT2 (p<=0.001). Moderate inter-reader agreement was observed for edema detection on T2w-FS images (ƙ=0.43), decreasing to fair on VirtuT2 images (ƙ=0.36). Conclusion: The 2D-U-Net can technically generate virtual T2w-FS images with high similarity to real T2w-FS images, though blurring remains a limitation. Further investigation of other architectures and using larger datasets are needed to improve clinical applicability.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"26 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503250","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-25DOI: 10.1101/2024.06.25.24309472
Lars Edenbrandt
Researchers in the RECOMIA network have been developing AI tools for the automated analysis of PET/CT studies in lymphoma patients. To enhance these AI tools, the CALGB 50303 dataset from The Cancer Imaging Archive was identified for inclusion in their project. Ensuring the quality of databases used for AI training is crucial, and one quality control (QC) measure involves the AI-based Fingerprint method to verify correct de-identification of clinical trial images. The study applied the Fingerprint method to PET/CT studies from 130 patients, successfully detecting an incorrectly de-identified study and identifying its correct trial identification number. This demonstrates the feasibility of using AI for QC in clinical trials. AI-based methods offer significant opportunities for enhancing QC, providing automated, consistent, and objective analyses that reduce the workload on human annotators. Integrating AI into QC processes promises to improve accuracy, consistency, and efficiency, thereby enhancing data integrity and the reliability of clinical trial results. This study underscores the importance of further developing AI-based QC methods in clinical trials.
{"title":"Fingerprint method applied to data from a phase III clinical trial","authors":"Lars Edenbrandt","doi":"10.1101/2024.06.25.24309472","DOIUrl":"https://doi.org/10.1101/2024.06.25.24309472","url":null,"abstract":"Researchers in the RECOMIA network have been developing AI tools for the automated analysis of PET/CT studies in lymphoma patients. To enhance these AI tools, the CALGB 50303 dataset from The Cancer Imaging Archive was identified for inclusion in their project. Ensuring the quality of databases used for AI training is crucial, and one quality control (QC) measure involves the AI-based Fingerprint method to verify correct de-identification of clinical trial images. The study applied the Fingerprint method to PET/CT studies from 130 patients, successfully detecting an incorrectly de-identified study and identifying its correct trial identification number. This demonstrates the feasibility of using AI for QC in clinical trials. AI-based methods offer significant opportunities for enhancing QC, providing automated, consistent, and objective analyses that reduce the workload on human annotators. Integrating AI into QC processes promises to improve accuracy, consistency, and efficiency, thereby enhancing data integrity and the reliability of clinical trial results. This study underscores the importance of further developing AI-based QC methods in clinical trials.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"72 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503251","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-25DOI: 10.1101/2024.06.25.24309247
Turay Cesur, Yasin Celal Gunes, Eren Camur, Mustafa Dağlı
Purpose This study evaluated the diagnostic accuracy and differential diagnosis capabilities of 12 Large Language Models (LLMs), one cardiac radiologist, and three general radiologists in cardiac radiology. The impact of ChatGPT-4o assistance on radiologist performance was also investigated.
{"title":"Empowering Radiologists with ChatGPT-4o: Comparative Evaluation of Large Language Models and Radiologists in Cardiac Cases","authors":"Turay Cesur, Yasin Celal Gunes, Eren Camur, Mustafa Dağlı","doi":"10.1101/2024.06.25.24309247","DOIUrl":"https://doi.org/10.1101/2024.06.25.24309247","url":null,"abstract":"<strong>Purpose</strong> This study evaluated the diagnostic accuracy and differential diagnosis capabilities of 12 Large Language Models (LLMs), one cardiac radiologist, and three general radiologists in cardiac radiology. The impact of ChatGPT-4o assistance on radiologist performance was also investigated.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"28 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503252","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-24DOI: 10.1101/2024.06.23.24309362
Hung-Ching Chang, Yusi Fang, Michael T. Gorczyca, Kayhan Batmanghelich, George C. Tseng
Causal mediation analysis provides a systematic approach to explore the causal role of one or more mediators in the association between exposure and outcome. In omics or imaging data analysis, mediators are often high-dimensional, which brings new statistical challenges. Existing methods either violate causal assumptions or fail in interpretable variable selection. Additionally, mediators are often highly correlated, presenting difficulties in selecting and prioritizing top mediators. To address these issues, we develop a framework using Partial Sum Statistic and Sample Splitting Strategy, namely PS5, for high-dimensional causal mediation analysis. The method provides a powerful global mediation test satisfying causal assumptions, followed by an algorithm to select and prioritize active mediators with quantification of individual mediation contributions. We demonstrate its accurate type I error control, superior statistical power, reduced bias in mediation effect estimation, and accurate mediator selection using extensive simulations of varying levels of effect size, signal sparsity, and mediator correlations. Finally, we apply PS5 to an imaging genetics dataset of chronic obstructive pulmonary disease (COPD) patients (N=8,897) in the COPDGene study to examine the causal mediation role of lung images (p=5,810) in the associations between polygenic risk score and lung function and between smoking exposure and lung function, respectively. Both causal mediation analyses successfully estimate the global indirect effect and detect mediating image regions. Collectively, we find a region in the lower lobe of the right lung with a strong and concordant mediation effect for both genetic and environmental exposures. This suggests that targeted treatment toward this region might mitigate the severity of COPD due to genetic and smoking effects.
因果中介分析提供了一种系统方法，用于探索一个或多个中介因素在暴露与结果之间的关联中的因果作用。在omics或成像数据分析中，中介因子通常是高维的，这给统计带来了新的挑战。现有的方法要么违反因果假设，要么在可解释的变量选择上失败。此外，中介因子往往高度相关，这给选择和优先考虑顶级中介因子带来了困难。为了解决这些问题，我们开发了一种使用偏和统计和样本分割策略（即 PS5）进行高维因果中介分析的框架。该方法提供了一个满足因果假设的强大的全局中介检验，随后提供了一种算法来选择和优先考虑活跃的中介，并量化单个中介的贡献。我们通过对不同水平的效应大小、信号稀疏性和中介相关性进行大量模拟，证明了该方法具有准确的 I 类误差控制、出色的统计能力、降低中介效应估计偏差以及准确的中介选择。最后，我们将 PS5 应用于 COPDGene 研究中慢性阻塞性肺病（COPD）患者（N=8897）的影像遗传学数据集，分别研究了肺部影像（p=5810）在多基因风险评分与肺功能之间以及吸烟暴露与肺功能之间的因果中介作用。这两项因果中介分析都成功地估算出了整体间接效应，并发现了中介图像区域。总之，我们发现右肺下叶的一个区域对遗传和环境暴露具有强烈且一致的中介效应。这表明，针对该区域的针对性治疗可能会减轻慢性阻塞性肺病因遗传和吸烟影响而导致的严重程度。
{"title":"High-dimensional causal mediation analysis by partial sum statistic and sample splitting strategy in imaging genetics application","authors":"Hung-Ching Chang, Yusi Fang, Michael T. Gorczyca, Kayhan Batmanghelich, George C. Tseng","doi":"10.1101/2024.06.23.24309362","DOIUrl":"https://doi.org/10.1101/2024.06.23.24309362","url":null,"abstract":"Causal mediation analysis provides a systematic approach to explore the causal role of one or more mediators in the association between exposure and outcome. In omics or imaging data analysis, mediators are often high-dimensional, which brings new statistical challenges. Existing methods either violate causal assumptions or fail in interpretable variable selection. Additionally, mediators are often highly correlated, presenting difficulties in selecting and prioritizing top mediators. To address these issues, we develop a framework using Partial Sum Statistic and Sample Splitting Strategy, namely PS5, for high-dimensional causal mediation analysis. The method provides a powerful global mediation test satisfying causal assumptions, followed by an algorithm to select and prioritize active mediators with quantification of individual mediation contributions. We demonstrate its accurate type I error control, superior statistical power, reduced bias in mediation effect estimation, and accurate mediator selection using extensive simulations of varying levels of effect size, signal sparsity, and mediator correlations. Finally, we apply PS5 to an imaging genetics dataset of chronic obstructive pulmonary disease (COPD) patients (<em>N</em>=8,897) in the COPDGene study to examine the causal mediation role of lung images (<em>p</em>=5,810) in the associations between polygenic risk score and lung function and between smoking exposure and lung function, respectively. Both causal mediation analyses successfully estimate the global indirect effect and detect mediating image regions. Collectively, we find a region in the lower lobe of the right lung with a strong and concordant mediation effect for both genetic and environmental exposures. This suggests that targeted treatment toward this region might mitigate the severity of COPD due to genetic and smoking effects.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"106 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141503253","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：