Predicting Conversion of Mild Cognitive Impairments to Alzheimer’s Disease and Exploring Impact of Neuroimaging

Abstract

Nowadays, a lot of scientific efforts are concentrated on the diagnosis of Alzheimers Disease (AD) applying deep learning methods to neuroimaging data. Even for 2017, there were published more than a hundred papers dedicated to AD diagnosis, whereas only a few works considered a problem of mild cognitive impairments (MCI) conversion to AD. However, the conversion prediction is an important problem as approximately 15% of patients with MCI converges to AD every year. In the current work, we are focusing on the conversion prediction using brain Magnetic Resonance Imaging and clinical data, such as demographics, cognitive assessments, genetic, and biochemical markers. First of all, we applied state-of-the-art deep learning algorithms on the neuroimaging data and compared these results with two machine learning algorithms that we fit using the clinical data. As a result, the models trained on the clinical data outperform the deep learning algorithms applied to the MR images. To explore the impact of neuroimaging further, we trained a deep feed-forward embedding using similarity learning with Histogram loss on all available MRIs and obtained 64-dimensional vector representation of neuroimaging data. The use of learned representation from the deep embedding allowed to increase the quality of prediction based on the neuroimaging. Finally, the current results on this dataset show that the neuroimaging does affect conversion prediction, however, cannot noticeably increase the quality of the prediction. The best results of predicting MCI-to-AD conversion are provided by XGBoost algorithm trained on the clinical and embedding data. The resulting accuracy is ACC = 0.76±0.01 and the area under the ROC curve -ROC AUC = 0.86 ± 0.01.

Deep learning prediction of mild cognitive impairment conversion to Alzheimer’s disease at 3 years after diagnosis using longitudinal and whole-brain 3D MRI

“…We employed 3D MRI instead of 2D multi-slice MRI. Previous studies have also reported MCI to AD prediction using a sequential full volume 3D architecture have obtained BAs of 0.75 (Basaia et al, 2019) and0.73 (Wen et al, 2020), while a study using residual architecture showed a resulting BA of 0.67 (Shmulev & Belyaev, 2018) but did not involve longitudinal MRI data. Some studies used predetermined 3D patches uniformly sampled across the brain (Lian et al, 2018;Liu et al, 2018;Wen et al, 2020).…”

“…Deep learning classification amongst normal cognition (NC), MCI and AD based on magnetic resonance imaging (MRI) data have been reported ( Cheng et al, 2017 ; Korolev et al, 2017 ; Wen et al, 2020 ). By contrast, there are comparatively fewer studies that reported prediction of MCI to AD conversion using deep learning of MRI data ( Lian et al, 2018 ; Lin et al, 2018 ; Liu et al, 2018 ; Shmulev & Belyaev, 2018 ; Basaia et al, 2019 ; Wen et al, 2020 ). A few ML studies used extracted brain structures or cortical thicknesses, and some used 3D patches from predetermined locations across the brain, but not whole-brain MRI data, to predict MCI to AD conversion ( Lian et al, 2018 ; Liu et al, 2018 ; Wen et al, 2020 ).…”

Deep learning for brain disorders: from data processing to disease treatment

“…Studies in this section aim at predicting the progression of patients in a pre-defined time frame. The length of this time frame depends on the disease studied: a few minutes for epilepsy [189] to several years for multiple sclerosis [190] or Alzheimer's disease [191][192][193][194][195].…”

“…Therefore, in this section we will only focus on transfer learning between these two tasks. Four possibilities were explored: i) transferring weights from AD versus HC and then fine-tuning on pMCI versus sMCI [191,192]; ii) transferring from a classification with the first class mixing AD and pMCI labels and the second class HC and sMCI labels, and then fine-tuning on pMCI versus sMCI [193]; iii) directly testing the network on pMCI and sMCI patients without fine-tuning after learning the AD versus HC classification [195]; and iv) directly learning sMCI versus pMCI classification [194]. It is important to note that the definition of the labels may vary between studies.…”

Peer Review #1 of "Deep learning prediction of mild cognitive impairment conversion to Alzheimer’s disease at 3 years after diagnosis using longitudinal and whole-brain 3D MRI (v0.1)"

“…Deep learning classification amongst normal cognition (NC), MCI and AD based on magnetic resonance imaging (MRI) data have been reported (Cheng et al 2017;Korolev et al 2017;Wen et al 2020). By contrast, there are comparatively fewer studies that reported prediction of MCI to AD conversion using deep learning of MRI data (Lian et al 2018;Lin et al 2018;Liu et al 2018;Shmulev & Belyaev 2018;Basaia et al 2019;Wen et al 2020). A few ML studies used extracted brain structures or cortical thicknesses, and some used 3D patches from predetermined locations across the brain, but not whole-brain MRI data, to predict MCI to AD conversion (Lian et al 2018;Liu et al 2018;Wen et al 2020).…”