Contributions
Abstract: 234
Type: Free Communications
Abstract Category: Pathology and pathogenesis of MS - MRI and PET
Introduction: Recently, deep learning approaches, in particular convolutional neural networks (CNNs), have led to state-of-the-art results in a wide range of medical imaging tasks including lesion segmentation and disease detection. By using a hierarchical architecture and not relying on hand-crafted or predefined features such as lesion volume, CNNs have the potential to make use of latent data characteristics not primarily associated with a certain disease.
Objectives: In the present study, we evaluate the ability of CNNs to separate structural magnetic resonance imaging (MRI) data of clinically definite multiple sclerosis (CDMS) patients and healthy controls.
Aims: The aim of this study is the comparison of MRI-based CNNs and lesion volume for identifying CDMS patients.
Methods: Fluid-attenuated inversion recovery (FLAIR) volumes of 76 CDMS patients and 71 healthy controls entered the analysis. After registering the volumes (rigid-body) to an MNI template, a CNN of 4 convolutional layers and 3 fully-connected layers was trained on the binary classification task. To prevent overfitting the model was regularized using dropout, batch normalization and early stopping. For reference, we performed a classification analysis using lesion volume and a linear support vector machine. Lesion volume was calculated based on manually segmented lesion masks available for all subjects in this study. To ensure generalizability of our results, we performed a 7-fold cross-validation and report mean accuracies and standard deviations (std).
Results: Using CNNs, we obtained a classification accuracy of 82.31 % (std 7.93 %) for CMDS patients and healthy controls. Based on lesion volume, we attained an accuracy of 73.13 % (std 9.08 %).
Conclusions: By showing that CNNs outperform the standard MS marker of lesion volume, we demonstrate the great potential of CNNs in characterizing structural MRI data in MS. Further studies are needed to understand what features are actually driving the classification decision and if these CNN representations are also useful in predicting disease progression.
Disclosure: FE, ES, RG, MW, JDH and KR (Ritter) have nothing to disclose.
JBS has received travel grants and speaking fees from Bayer Healthcare, Biogen Idec, Merck Serono, sanofi-aventis/Genzyme, Teva Pharmaceuticals, and Novartis unrelated to the presented work. AUB is cofounder and shareholder of Motognosis and Nocturne. He is named as inventor on several patent applications regarding MS serum biomarkers, OCT image analysis and perceptive visual computing. KR (Ruprecht) was supported by the German Ministry of Education and Research (BMBF/KKNMS, Competence Network Multiple Sclerosis) and has received research support from Novartis and Merck Serono as well as speaking fees and travel grants from Guthy Jackson Charitable Foundation, Bayer Healthcare, Biogen Idec, Merck Serono, sanofi-aventis/Genzyme, Teva Pharmaceuticals, Roche and Novartis. JK received congress registration fee from Biogen and research support from Krankheitsbezogenes Kompetenznetz Multiple Sklerose. MS holds a patent for manufacturing of phantoms for computed tomography imaging with 3D printing technology and received research support from Federal Ministry of Economics and Technology. FP serves on the scientific advisory board for Novartis; received speaker honoraria and travel funding from Bayer, Novartis, Biogen Idec, Teva, Sanofi-Aventis/Genzyme, Merck Serono, Alexion, Chugai, MedImmune, and Shire; is an academic editor for PLoS ONE; is an associate editor for Neurology Neuroimmunology & Neuroinflammation; consulted for SanofiGenzyme, Biogen Idec, MedImmune, Shire, and Alexion; and received research support from Bayer, Novartis, Biogen Idec, Teva, Sanofi-Aventis/Genzyme, Alexion, Merck Serono, German Research Council, Werth Stiftung of the City of Cologne, German Ministry of Education and Research, Arthur Arnstein Stiftung Berlin, EU FP7 Framework Program, Arthur Arnstein Foundation Berlin, Guthy Jackson Charitable Foundation, and National Multiple Sclerosis of the USA.
Abstract: 234
Type: Free Communications
Abstract Category: Pathology and pathogenesis of MS - MRI and PET
Introduction: Recently, deep learning approaches, in particular convolutional neural networks (CNNs), have led to state-of-the-art results in a wide range of medical imaging tasks including lesion segmentation and disease detection. By using a hierarchical architecture and not relying on hand-crafted or predefined features such as lesion volume, CNNs have the potential to make use of latent data characteristics not primarily associated with a certain disease.
Objectives: In the present study, we evaluate the ability of CNNs to separate structural magnetic resonance imaging (MRI) data of clinically definite multiple sclerosis (CDMS) patients and healthy controls.
Aims: The aim of this study is the comparison of MRI-based CNNs and lesion volume for identifying CDMS patients.
Methods: Fluid-attenuated inversion recovery (FLAIR) volumes of 76 CDMS patients and 71 healthy controls entered the analysis. After registering the volumes (rigid-body) to an MNI template, a CNN of 4 convolutional layers and 3 fully-connected layers was trained on the binary classification task. To prevent overfitting the model was regularized using dropout, batch normalization and early stopping. For reference, we performed a classification analysis using lesion volume and a linear support vector machine. Lesion volume was calculated based on manually segmented lesion masks available for all subjects in this study. To ensure generalizability of our results, we performed a 7-fold cross-validation and report mean accuracies and standard deviations (std).
Results: Using CNNs, we obtained a classification accuracy of 82.31 % (std 7.93 %) for CMDS patients and healthy controls. Based on lesion volume, we attained an accuracy of 73.13 % (std 9.08 %).
Conclusions: By showing that CNNs outperform the standard MS marker of lesion volume, we demonstrate the great potential of CNNs in characterizing structural MRI data in MS. Further studies are needed to understand what features are actually driving the classification decision and if these CNN representations are also useful in predicting disease progression.
Disclosure: FE, ES, RG, MW, JDH and KR (Ritter) have nothing to disclose.
JBS has received travel grants and speaking fees from Bayer Healthcare, Biogen Idec, Merck Serono, sanofi-aventis/Genzyme, Teva Pharmaceuticals, and Novartis unrelated to the presented work. AUB is cofounder and shareholder of Motognosis and Nocturne. He is named as inventor on several patent applications regarding MS serum biomarkers, OCT image analysis and perceptive visual computing. KR (Ruprecht) was supported by the German Ministry of Education and Research (BMBF/KKNMS, Competence Network Multiple Sclerosis) and has received research support from Novartis and Merck Serono as well as speaking fees and travel grants from Guthy Jackson Charitable Foundation, Bayer Healthcare, Biogen Idec, Merck Serono, sanofi-aventis/Genzyme, Teva Pharmaceuticals, Roche and Novartis. JK received congress registration fee from Biogen and research support from Krankheitsbezogenes Kompetenznetz Multiple Sklerose. MS holds a patent for manufacturing of phantoms for computed tomography imaging with 3D printing technology and received research support from Federal Ministry of Economics and Technology. FP serves on the scientific advisory board for Novartis; received speaker honoraria and travel funding from Bayer, Novartis, Biogen Idec, Teva, Sanofi-Aventis/Genzyme, Merck Serono, Alexion, Chugai, MedImmune, and Shire; is an academic editor for PLoS ONE; is an associate editor for Neurology Neuroimmunology & Neuroinflammation; consulted for SanofiGenzyme, Biogen Idec, MedImmune, Shire, and Alexion; and received research support from Bayer, Novartis, Biogen Idec, Teva, Sanofi-Aventis/Genzyme, Alexion, Merck Serono, German Research Council, Werth Stiftung of the City of Cologne, German Ministry of Education and Research, Arthur Arnstein Stiftung Berlin, EU FP7 Framework Program, Arthur Arnstein Foundation Berlin, Guthy Jackson Charitable Foundation, and National Multiple Sclerosis of the USA.