Computational Radiology & Artificial Intelligence unit, Division of Radiology and Nuclear Medicine, Oslo University Hospital, 0372, Oslo, Norway; Center for Lifespan Changes in Brain and Cognition, Department of Psychology, University of Oslo, 0373, Oslo, Norway.
Center for Lifespan Changes in Brain and Cognition, Department of Psychology, University of Oslo, 0373, Oslo, Norway; Norwegian Centre for Mental Disorders Research, Oslo University Hospital & Institute of Clinical Medicine, University of Oslo, 0373, Oslo, Norway.
J Neurosci Methods. 2024 Nov;411:110253. doi: 10.1016/j.jneumeth.2024.110253. Epub 2024 Aug 20.
There is a broad interest in deploying deep learning-based classification algorithms to identify individuals with Alzheimer's disease (AD) from healthy controls (HC) based on neuroimaging data, such as T1-weighted Magnetic Resonance Imaging (MRI). The goal of the current study is to investigate whether modern, flexible architectures such as EfficientNet provide any performance boost over more standard architectures.
MRI data was sourced from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and processed with a minimal preprocessing pipeline. Among the various architectures tested, the minimal 3D convolutional neural network SFCN stood out, composed solely of 3x3x3 convolution, batch normalization, ReLU, and max-pooling. We also examined the influence of scale on performance, testing SFCN versions with trainable parameters ranging from 720 up to 2.9 million.
SFCN achieves a test ROC AUC of 96.0% while EfficientNet got an ROC AUC of 94.9 %. SFCN retained high performance down to 720 trainable parameters, achieving an ROC AUC of 91.4%.
The SFCN is compared to DenseNet and EfficientNet as well as the results of other publications in the field.
The results indicate that using the minimal 3D convolutional neural network SFCN with a minimal preprocessing pipeline can achieve competitive performance in AD classification, challenging the necessity of employing more complex architectures with a larger number of parameters. This finding supports the efficiency of simpler deep learning models for neuroimaging-based AD diagnosis, potentially aiding in better understanding and diagnosing Alzheimer's disease.
基于神经影像学数据（如 T1 加权磁共振成像（MRI）），利用深度学习分类算法从健康对照（HC）中识别阿尔茨海默病（AD）患者，这一方法引起了广泛关注。本研究旨在探究现代灵活的架构（如 EfficientNet）是否比更标准的架构提供任何性能提升。
从阿尔茨海默病神经影像学倡议（ADNI）获取 MRI 数据，并通过最小预处理流水线进行处理。在测试的各种架构中，突出的是最小的 3D 卷积神经网络 SFCN，它仅由 3x3x3 卷积、批量归一化、ReLU 和最大池化组成。我们还研究了规模对性能的影响，测试了具有从 720 到 290 万可训练参数的 SFCN 版本。
SFCN 的测试 ROC AUC 为 96.0%，而 EfficientNet 的 ROC AUC 为 94.9%。SFCN 在可训练参数低至 720 时仍保持高性能，其 ROC AUC 为 91.4%。
将 SFCN 与 DenseNet 和 EfficientNet 进行比较，并与该领域的其他出版物的结果进行比较。
结果表明，使用具有最小预处理流水线的最小 3D 卷积神经网络 SFCN 可以在 AD 分类中实现有竞争力的性能，这质疑了使用具有更多参数的更复杂架构的必要性。这一发现支持了用于神经影像学 AD 诊断的更简单深度学习模型的效率，可能有助于更好地理解和诊断阿尔茨海默病。
