Deep Learning-Based Classification and Voxel-Based Visualization of Frontotemporal Dementia and Alzheimer’s Disease

Abstract

Frontotemporal dementia (FTD) and Alzheimer’s disease (AD) have overlapping symptoms, and accurate differential diagnosis is important for targeted intervention and treatment. Previous studies suggest that the deep learning (DL) techniques have the potential to solve the differential diagnosis problem of FTD, AD and normal controls (NCs), but its performance is still unclear. In addition, existing DL-assisted diagnostic studies still rely on hypothesis-based expert-level preprocessing. On the one hand, it imposes high requirements on clinicians and data themselves; On the other hand, it hinders the backtracking of classification results to the original image data, resulting in the classification results cannot be interpreted intuitively. In the current study, a large cohort of 3D T1-weighted structural magnetic resonance imaging (MRI) volumes (n = 4,099) was collected from two publicly available databases, i.e., the ADNI and the NIFD. We trained a DL-based network directly based on raw T1 images to classify FTD, AD and corresponding NCs. And we evaluated the convergence speed, differential diagnosis ability, robustness and generalizability under nine scenarios. The proposed network yielded an accuracy of 91.83% based on the most common T1-weighted sequence [magnetization-prepared rapid acquisition with gradient echo (MPRAGE)]. The knowledge learned by the DL network through multiple classification tasks can also be used to solve subproblems, and the knowledge is generalizable and not limited to a specified dataset. Furthermore, we applied a gradient visualization algorithm based on guided backpropagation to calculate the contribution graph, which tells us intuitively why the DL-based networks make each decision. The regions making valuable contributions to FTD were more widespread in the right frontal white matter regions, while the left temporal, bilateral inferior frontal and parahippocampal regions were contributors to the classification of AD. Our results demonstrated that DL-based networks have the ability to solve the enigma of differential diagnosis of diseases without any hypothesis-based preprocessing. Moreover, they may mine the potential patterns that may be different from human clinicians, which may provide new insight into the understanding of FTD and AD.

Predictive Modelling of Brain Disorders with Magnetic Resonance Imaging: A Systematic Review of Modelling Practices, Transparency, and Interpretability in the use of Convolutional Neural Networks

“…This highlights a lack of methodological transparency across the considered research, especially considering the many different subjective choices required during model construction that can have misleading effects on the overall performance of the system. Of the studies that did make code available, no paper provided detailed tutorials of preprocessing and model construction -understandably, the quality and thoroughness of reported code is another important aspect of reproducibility and transparency which is not solved by making code available [55,57,58,59,60,48]. Additionally, most papers considered were from journals, meaning that the majority of studies underwent some form of peer review (43/55)…”

“…The 9 aforementioned studies did not provide any information on implementation details. Of the remaining 10 studies, 5 provided code, but as previously mentioned did not include detailed walkthroughs of their interpretation pipelines [57,48,58,55,60]. A subsection of studies also specifically underscored that extensive, expert driven preprocessing is not required with deep learning studies, and almost all studies alluded to this fact in their introductions [60,69,79,51,52].…”

“…Of the remaining 10 studies, 5 provided code, but as previously mentioned did not include detailed walkthroughs of their interpretation pipelines [57,48,58,55,60]. A subsection of studies also specifically underscored that extensive, expert driven preprocessing is not required with deep learning studies, and almost all studies alluded to this fact in their introductions [60,69,79,51,52]. This position downplays the importance of expert opinion in model interpretation and preprocessing decisions for medical imaging studies.…”

“…This position downplays the importance of expert opinion in model interpretation and preprocessing decisions for medical imaging studies. Without explicit knowledge concerning what image aspects to exclude, researchers can include irrelevant information during model training, as demonstrated by the inclusion of skull and neck information in several studies [60,91,38,84,67,77]. In practice, this would mean the models may have picked up on irrelevant information about neck size or skull thickness, that, if used in clinical applications, could lead to misclassifications of patients with those specific physical characteristics, which may have nothing to do with the condition of interest.…”

“…In practice, this would mean the models may have picked up on irrelevant information about neck size or skull thickness, that, if used in clinical applications, could lead to misclassifications of patients with those specific physical characteristics, which may have nothing to do with the condition of interest. Furthermore, certain studies avoided considering interpretability in greater detail due to their lack of expertise, which implies that expert knowledge is a requirement for thorough considerations of saliency [60]. These examples illustrate the crucial importance of both domain knowledge and interpretability, which in this case would have highlighted potentially spurious relationships the model may have been using during classification.…”