Alzheimer's Disease (AD), Parkinson's Disease (PD) and SchiZophrenia (SZ) are the principal brain dysfunctions, whose diagnosis is still challenging due to their complex clinical manifestations. Structural Magnetic Resonance (MR) is an effective way to diagnose brain dysfunctions by catching structural (like the ones caused by AD and PD) or connectivity (like the ones caused by SZ) alterations. As far as we know, we were the first to propose an end-to-end Convolutional Long Short-Term Memory (ConvLSTM)-based framework, ConvLSTM4AD, and to develop an improved version of it, Brain-on-Cloud, for AD automatic recognition from MR whole-brain scans. This research aims to extend their applicability to other brain dysfunctions and verify their efficiency through the development of CASPAR, a cloud-based AD, PD and SZ automatic recognizer that employs a neural network, whose principal layer is a ConvLSTM layer, fed with MR whole-brain scans. To this aim, we selected unenhanced MR scans from the ADNI database, the PPMI database and the COBRE respectively for AD, PD and SZ, and preprocessed them. Next, we divided the preprocessed MR scans in train, validation and test sets, and performed train data augmentation. Then, we fed the preprocessed MR scans to the ConvLSTM-based neural network for automatic feature extraction and classification on cloud. Eventually, we evaluated the neural network performance on test data. CASPAR achieved very promising results in automatically recognizing AD (average sensitivity of 96.43%), PD (average sensitivity of 95.48%) and SZ (average sensitivity of 100.00%), even outperforming recent state-of-the-art frameworks. With the development of CASPAR, we demonstrated that a cloud-based and ConvLSTM-based framework that works well for AD automatic recognition, works well also for PD and SZ automatic recognition, paving the way to a fully-integrated system for simultaneous automatic recognition of multiple brain dysfunctions from MR whole-brain scans.
CASPAR: Cloud-based Alzheimer's, Schizophrenia and Parkinson's Automatic Recognizer / Tomassini, S.; Sernani, P.; Falcionelli, N.; Dragoni, A. F.. - ELETTRONICO. - (2022), pp. 6-10. (Intervento presentato al convegno 1st IEEE International Workshop on Metrology for Extended Reality, Artificial Intelligence and Neural Engineering, MetroXRAINE 2022 tenutosi a Roma nel 2022) [10.1109/MetroXRAINE54828.2022.9967634].
CASPAR: Cloud-based Alzheimer's, Schizophrenia and Parkinson's Automatic Recognizer
Tomassini S.
;Sernani P.;Falcionelli N.;Dragoni A. F.
2022-01-01
Abstract
Alzheimer's Disease (AD), Parkinson's Disease (PD) and SchiZophrenia (SZ) are the principal brain dysfunctions, whose diagnosis is still challenging due to their complex clinical manifestations. Structural Magnetic Resonance (MR) is an effective way to diagnose brain dysfunctions by catching structural (like the ones caused by AD and PD) or connectivity (like the ones caused by SZ) alterations. As far as we know, we were the first to propose an end-to-end Convolutional Long Short-Term Memory (ConvLSTM)-based framework, ConvLSTM4AD, and to develop an improved version of it, Brain-on-Cloud, for AD automatic recognition from MR whole-brain scans. This research aims to extend their applicability to other brain dysfunctions and verify their efficiency through the development of CASPAR, a cloud-based AD, PD and SZ automatic recognizer that employs a neural network, whose principal layer is a ConvLSTM layer, fed with MR whole-brain scans. To this aim, we selected unenhanced MR scans from the ADNI database, the PPMI database and the COBRE respectively for AD, PD and SZ, and preprocessed them. Next, we divided the preprocessed MR scans in train, validation and test sets, and performed train data augmentation. Then, we fed the preprocessed MR scans to the ConvLSTM-based neural network for automatic feature extraction and classification on cloud. Eventually, we evaluated the neural network performance on test data. CASPAR achieved very promising results in automatically recognizing AD (average sensitivity of 96.43%), PD (average sensitivity of 95.48%) and SZ (average sensitivity of 100.00%), even outperforming recent state-of-the-art frameworks. With the development of CASPAR, we demonstrated that a cloud-based and ConvLSTM-based framework that works well for AD automatic recognition, works well also for PD and SZ automatic recognition, paving the way to a fully-integrated system for simultaneous automatic recognition of multiple brain dysfunctions from MR whole-brain scans.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
