Multimodal Explainability Using Class Activation Maps and Canonical Correlation for MI-EEG Deep Learning Classification

1. Introduction

The 2021 UNESCO Engineering for Sustainable Development report points out the role the profession has on the 2030 Agenda. The third goal, "ensure healthy lives and promote well-being for all at all ages," highlights the advancements in improving medical diagnosis and care through low-cost tools [1]. The most affordable tool in neuroscience is Electroencephalography (EEG), which is also the most studied thanks to its high temporal resolution and portability. EEG captures subjects’ neural bioelectric activity generated by neuron activation through electrodes placed on the scalp [2]. Techniques like Event-Related Potentials (ERPs) extract representative time-frequency information within the time-series data for understanding the neurological activity of a subject or a group [3]. Also, Brain-Computer Interfaces (BCI) profit from those patterns for controlling external devices, such as prostheses [4]. To learn the brain patterns, a BCI paradigm presents stimuli and asks the subject to perform tasks. For instance, the Motor Imagery (MI) paradigm, which consists of the mental rehearsal of motor tasks without physical movement, has been used to support the diagnosis, treatment, and follow-up of brain diseases [5].

However, EEG is far from the panacea for all neuroimaging needs. Due to its non-invasive nature, superficial EEG sacrifices spatial resolution and becomes highly susceptible to electromagnetic artifacts (e.g., electrical devices) and physiological noise (e.g., eye movement and muscle activity), yielding less useful features [6,7]. Moreover, the volume conduction effect introduces noise or cross-talk between electrodes, hampering the source localization of brain activity and the interpretation of EEG data [8]. Lastly, both between- and within-subject variability challenge the development of a universal MI-EEG algorithm. Differences in genetic, cognitive, and neurodevelopmental factors cause the same task or stimuli to evoke distinct brain patterns across individuals, complicating the creation of subject-independent solutions [9,10]. Additionally, as users become more familiar with the BCI device or task over time, their performance and initial brain patterns evolve, demanding personalized calibration sessions [11].

Traditional MI-EEG algorithms attempt to extract features by analyzing the signal’s power levels. The baseline Common Spatial Patterns (CSP) exploits the power distribution over the scalp to find spatial patterns discriminating two tasks [12]. Variations of CSP include L1-CSP, which regularizes the patterns using the L1-norm; Sparse Filter Band CSP (SFBCSP), which automatically selects useful spectral bands from a precomputed set [13]; and Multi-Kernel Stein Spatial Patterns (MKSSP), which extracts nonlinear patterns in a low-dimensional Riemannian manifold [14]. However, a low signal-to-noise ratio makes CSP and its variants extract features from artifacts rather than the actual EEG [15].

Conversely, machine learning algorithms exploit their ability to automatically learn features from a given training dataset to minimize the prediction error as much as possible [16]. Deep Learning (DL) methods extract nonlinear patterns, dealing with EEG noise and volume conduction effect [17]. Convolutional Neural Networks (CNNs), a DL model, are the most successful EEG feature extraction architectures because they look for space and time patterns [18]. Examples of CNNs for MI-EEG classification are EEGNet, ShallowConvNet, and DeepConvNet, which, like SFBCSP, also extract spatial patterns from certain frequency bands [19]. Unlike SFBCSP, the above architectures unravel nonlinear, deeper, and more complex patterns [20]. Other DL models for EEG applications include Autoencoders that embed EEG signals into a generative noise-reduced feature space [21]; Recurrent Neural Networks (RNNs) that exploit the sequential nature of EEG features [22]; and, more recently, Transformers that use their long-term memory to capture both global and local patterns [23].

However, there are two main concerns about the medical applications of the above DL models. Firstly, they become "black boxes," lacking interpretable information to understand each subject’s neurological abilities [24]. Secondly, they ignore the closed link between neurological, physiological, and personal behaviors [25]. Analyzing how these factors influence the model provides users and scientists with valuable information to improve the results beforehand. Thus, multimodal and multidomain strategies can couple information from patients’ moods and habits to understand their neurophysiological responses and exploit this additional information to improve model performance and interpretability.

This work proposes Multimodal and Explainable Deep Learning (MEDL) as an approach for MI-EEG discrimination and physiological interpretability. Specifically, our proposal is threefold: i) Different DL models are tested for subject-dependent MI-EEG discrimination. ii) A CAM-based approach is used to quantify and visualize relevant MI-EEG features. iii) A Questionnaire-MI Performance Canonical Correlation Analysis (QMIP-CCA) strategy is introduced as a multidomain explainability stage for physiological and MI-EEG discrimination non-linear feature matching. Experiments are carried out with the GIGAScience MI dataset due to its relatively large number of subjects and the additional questionnaire that it offers for physiological subject information [26]. The results obtained demonstrate that shallow networks achieve acceptable MI discrimination results. In addition, our CAM-based method allows us to code MI spatio-frequency group patterns and measure EEG features of the sensorimotor cortex in people who are getting better at MI. Finally, our QMIP-CCA allows quantifying and visualizing relevant physiological questions from tabular data and matching them to MI-EEG performance measures.

The agenda for this paper is as follows. Section 2 summarizes the related work. Section 3 describes the materials and methods. Section 4 describes the experiments and discuss the results. Lastly, Section 5 outlines the conclusions and future work.

2. Related work

Since Koles et al.’s work in 1990, CSP has been the go-to tool for feature extraction in EEG data [27]. CSP provides spatial filters that maximize the variance ratio of two multivariate signals, enabling the discrimination of two classes for classification purposes. Unfortunately, since the technique depends on the signals’ variance, it becomes sensitive to noise and struggles in small datasets [28]. In turn, variants of CSP have sprung up to enhance the original algorithm. L1CSP redefines the base objective in terms of the L1-norm instead of the usual L2-norm to reduce artifacts’ influence on the signal [29]. In contrast, FBCSP adds a bandpass filtering stage for multiple, manually selected frequency bands before the usual spatial filtering. After that, it uses a mutual information algorithm to select discriminant features [30]. Lastly, SFBCSP selects the bandpass filters semi-automatically, rather than manually like FBCSP, by integrating a sparse regression model to learn the optimal features from each input filter band [31]. However, these models remain power-reliant and, in the case of FBCSP and SFBCSP, require some form of manual input. Once features have been extracted, classification can be performed by various methods. One such technique are Support Vector Machines (SVMs). SVM works by finding the optimal hyperplane for separating classes [32]. However, SVMs fail to classify data when features are too similar between classes [33].

Thanks to their ability to recognize and extract non-linear features from raw EEG data [34], DL models solve many of these traditional algorithms’ shortcomings. CNNs are a family of DL strategies that scan the input signal for representative patterns, starting from simple structures until reaching a more complex combination of these initial features. These algorithms are commonly used for image processing, but by tuning the size and number of filters, it is possible to extract information from the temporal, spatial, and frequency domains, e.g., for MI-EEG classification. Also, unlike CSP, these models automatically fine-tune the best filters for the given task through gradient decent [35]. Some relevant examples of CNNs for MI-EEG include the EEGNet, KREEGNet, ShallowConvNet, DeepConvNet, TCFusionnet, and KCS-FCNet.

In particular, EEGNet gets different features from EEG data in three steps: temporal convolution, depthwise convolution (to find patterns in time and space), and separable convolution (to combine the data from the first two steps) [36]. Variations of this original architecture have since emerged. TCFusionNet, for example, integrates the EEGNet architecture with residual blocks composed of dilated convolutions, which add these residual features to the ones from the separable convolution to increase the size of the receptive field while avoiding the exploding/vanishing gradient problem of deeper models [37]. ShallowConvNet works similarly to EEGNet but skips the separable convolution stage and uses regular convolution for spatial filtering. This reduces the number of parameters requiring training, leading to faster training and better interpretability, but at the cost of some performance. DeepConvNet, on the other hand, goes much further by adding a series of 2D convolutions that get bigger and bigger to pull information from the earlier stages and find complex structures [19]. Regarding Deep Kernel Learning (DKL) methods, KREEGNet computes the functional connectivities from the temporal convolution through a Gaussian similarity, alongside Delta Kernel for label outputs, all to use Central Kernel Alignment (CKA) as a regularizer [38]. Likewise, KCS-FCNet computes the functional connectivities from a temporal convolution through a Gaussian kernel, much like KREEGNet; however, it then relies on the measured connectivities as input for a Fully-Connected block to classify the MI data on a high-dimensional space [39].

Nevertheless, as computational power has increased, more powerful and complex algorithms have been developed to extract more information from EEG signals. Deep Belief Networks (DBNs) stack multiple Restricted Boltzmann Machines (RBMs), which learn to reconstruct a given input through unsupervised training; however, as a result of using RBMs, DBNs require a pre-training stage on a separate dataset, which is a luxury only available on larger databases [40]. Another set of neural networks for EEG data is Autoencoders, which have been used as dynamic Principal Component Analysis (PCAs) to select the most relevant characteristics before classification [41]. Ideally, the model will filter out any kind of noise contaminating the signals, as it acts as redundant information, leaving only features intrinsic to the EEG for classification [42]. Unfortunately, physiological noise is extremely difficult to properly filter, as it is influenced by factors such as stress levels, personal background, and even the testing environment [43]. Alternatively, RNNs are a natural choice for EEG analysis, as their ability to remember previous inputs makes them ideal for time series data, thanks to their strong modeling capabilities, making them useful at leveraging cross-series information, even when handling heterogenous signals [44]. Recently, Transformer networks utilize an Attention Mechanism to capture global information from EEG by encoding information from temporal, frequency, and spatial features, allowing the model to identify relevant sample dependencies while dealing with the low signal-noise ratio problem [45]. Despite these models’ multiple advantages, their use is severely limited in tasks requiring interpretability, as these architectures lack tools to allow users to properly assess the cause for a given output, as it is difficult to assure if their performance is thanks to them extracting meaningful information, or noise [46]. Furthermore, precisely due to their complexity, they are vulnerable to overfitting [47]. EEG-BCI classification methods are summarized in Figure 1.

Now, multiple strategies have emerged to provide interpretable results from DL-based methods. The authors in [48] propose four different types of explanations offered by the many algorithms: Example, Attribution, Hidden Semantics, and Rules. Example algorithms check which inputs are similar to each other according to the model. Attribution looks at which elements of the input had the greatest influence on the given output. Hidden Semantics employs the network’s neurons to explain the output; for instance, when classifying animals, it determines whether the neurons are focusing on the head, legs, or other parts of the creature. Finally, rules explain the output in terms of a series of decisions taken by the model; in their most simple incarnation, these rules take the form "If X is present, then Y." However, when working with EEG data, not every algorithm will provide useful insight. Rules as explanation reduce the model to a Decision Tree algorithm but do not necessarily create explainable rules themselves [49]. This leaves open the other three types of explanations. Hidden Semantics are useful for evaluating the importance of specific features at certain points, but they produce abstract results the deeper the neurons are [50]. For inter-subject analysis, Example algorithms are capable of finding similarities between subjects, providing information as to which patients share similar attributes, but struggle in providing explainability to extracted features as these must already make sense to the BCI professionals beforehand [51]. Next, Attribution possesses the most potential for BCI insight, as these techniques map the importance of the output back to the original input (EEG data). Common attribution algorithms are CAMs, which create a mask that highlights which pixels from the input image supply the most important information for the output. Introduced by [52], CAMs map out the elements within an image most relevant for the CNN. This method involves weighting the activation maps from a given layer and finding the average contribution of each pixel to the model’s decision. Grad-CAM, by [53], generalizes CAMs by redefining the weights in terms of the gradient produced by the model. However, this technique uses a global average for weight calculations, assuming each activation to be equally important. Grad-Cam++ [54] then redefines Grad-Cam to be in terms of a weighted average instead of global. Next, LayerCAM enables the use of CAMs for any convolutional layer [55]. By utilizing the backward class-specific gradients, it is possible to generate a separate weight for each spatial location.

Another strategy to improve EEG classification and interpretability is to incorporate information from different domains into the DL structure. Multi-domain Fusion Deep Graph CNN (MdGCNN) by [56] fuses time-frequency and spatial information through the use of graph convolutions, which learn the discriminant features across the domains, followed by a sort pooling layer to act as a fusion stage to bridge the extracted information to regular convolutions, which then produce the output. Still, this approach does not include information from external sources to the EEG, limiting its interpretability, unlike the authors in [57] who perform emotion recognition through RNNs using visual information from videos in addition to the EEG, fusing into a hierarchical attention mechanism to organize features based on their perceived significance. This method allows the analysis of how physiological and neurological responses relate to each other during classification. Lastly, [58] uses a deep and wide CNN to pull out features and perform an initial classification. Kernel matching via Gaussian embedding is then used to combine data from questionnaires and make the model output more accurate.

Ultimately, it is evident that traditional feature extraction algorithms, like CSP, provide the best interpretability but are also the least powerful for EEG-based classification, primarily due to their need for manual tuning. Complex DL solutions such as Autoencoder, RNNs, and Transformers offer the ability to extract much more information when compared to other solutions, but their results aren’t easily interpretable or require large datasets [59]. This, in turn, leaves CNNs-based EEGNet variants as the best compromise between the two. They work similarly to traditional CSP algorithms, requiring no manual tuning for the filters; can be expanded upon to exploit information present in much deeper structures; and possess extremely useful interpretation tools in the form of CAMs.

3. Materials and Methods

3.1. GIGAScience Dataset for MI-EEG

DB-I - GiGaScience[26] (http://gigadb.org/dataset/100295). It consists of 52 subjects, 50 of which have their EEG data available for evaluation. Each subject is asked to perform a single session of MI, comprised of five to six runs with 100 to 120 trials per class. Each trial lasts seven seconds, starting with a blank screen, followed by a cue within two seconds. When the cue appears on screen, the subject imagines moving their left or right hand. The trial ends with two more seconds of a blank screen and an inter-trial break of 0.1 to 0.8 seconds. The EEG data was collected using 64 electrodes placed according to the international 10-10 system, as seen on Figure 2, sampled at 512 Hz. Actual movement and six types of non-task-related data (blinking eyes, eyeball movement up/down, eyeball movement left/right, head movement, jaw clenching, and resting state) were also collected.

Additionally, subjects were asked to complete a physiological questionnaire during the MI experiment. Before beginning the MI experiment, subjects answered 15 personal questions such as Age, Sex, BCI experience, or any recent consumption of coffee, alcohol, or cigarettes. After every run, subjects answered another ten questions about their mood and expected performance. Finally, after the MI experiment, subjects had to answer four questions regarding their thoughts on the experiment and overall personal performance. In total, each subject answered 69 questions. Questions relating to how the patient felt were answered on a scale from one to five, while open questions such as age or expected performance were answered numerically without decimals. However, not all questions were utilized in this study. Shannon’s entropy, as presented in Equation 1:

$$H\left({\xi }_q\right)=-\mathbb{E}\left\{log\left(p\left({\xi }_{rq}\right)\right):\forall r\in R\right\},$$

(1)

was measured for each of the Q questions, ${\xi }_{rq}\in {\xi }_q$, $q\in Q$, correctly encoded and normalized between $[0,1]$, along the R subjects to find the ones that had the most expected information, with the response’s histogram giving an estimate of $p\left({\xi }_{rq}\right)$. Figure 3 illustrates a significant decline in entropy following the initial 31 questions, arranged from highest to lowest entropy, indicating that subsequent questions yield considerably less informational value. As a result, only questions with entropy above or equal to the 25th percentile were selected for this study. This then totals 52 out of the 69 available questions (see Table A1).

3.2. Subject-Dependent MI-EEG Classification Using Deep Learning

Let $\mathcal{D}={\{{\mathbf{X}}_n\in {\mathbb{R}}^{C\times \tau },{\mathbf{y}}_n\in {\{0,1\}}^{\tilde{K}}\}}_{n=1}^N$ be a subject-dependent input-output MI-EEG dataset, gathering N trials, $\tau$ time samples, C channels, and $\tilde{K}$ MI-classes ($\tilde{K}=2$ for the GIGAScience dataset). To optimize the meaningful EEG spatial-temporal-spectral patterns from a given $\mathbf{X}\in \mathcal{D}$ and diminish noise for enhanced MI class prediction, a DL approach can be employed as follows:

$$\widehat{\mathbf{y}}=\tilde{f}\left(\mathbf{X}\right|\theta )=(f_L\circ f_{L-1}\circ \cdots \circ f_1)\left(\mathbf{X}\right),$$

(2)

$\widehat{\mathbf{y}}\in {[0,1]}^K,$${\sum }_{k=1}^K{\widehat{y}}_k=1,$ L stands for the number of layers, and:

$${\mathbf{Z}}_l=f_l\left({\mathbf{Z}}_{l-1}\right)=\phi ({\mathbf{Z}}_{l-1}\otimes {\mathbf{W}}_l+{\mathbf{B}}_l),$$

(3)

where $\phi (·)$ is a given non-linear activation function, ${\mathbf{Z}}_l\in {\mathbb{R}}^{C_l^{\prime }\times {\tau }_l^{\prime }\times {\tilde{P}}_l}$ is the l-th feature map; ${\mathbf{W}}_l$ and ${\mathbf{B}}_l$ hold the weights and bias of proper size, ${\tilde{P}}_l$ stands for the number of filters, and $\theta ={\{{\mathbf{W}}_l,{\mathbf{b}}_l\}}_{l=1}^L$ collects the network parameters. Furthermore, ⊗ represents the tensor product operator for fully connected, convolutional, or recurrent layers.

Then, an optimization problem can be formulated as:

$$\widehat{\theta }=arg\underset{\theta }{min}\mathbb{E}\{\mathcal{L}({\mathbf{y}}_n,\tilde{f}\left({\mathbf{X}}_n\right|\theta )):\forall n\in \{1,2,\cdots ,N\}\},$$

(4)

where $\mathcal{L}(·,·)$ is a given loss function, i.e., cross-entropy. In addition, a gradient descent-based framework using back-propagation is employed to optimize the parameter set [60]:

$${\theta }_i={\theta }_{i-1}-{\eta }_i{\displaystyle \frac{\partial }{\partial {\theta }_i}}\left\{{\displaystyle \frac{1}{N}}\sum _{n=1}^N\mathcal{L}({\mathbf{y}}_n,\tilde{f}\left({\mathbf{X}}_n\right|{\theta }_i))\right\},$$

(5)

where a sample mean estimation is used to approximate the expected value, and an autodiff-based approach computes the gradient (${\eta }_i\in {\mathbb{R}}^{+}$ is a learning rate).

Here, the following MI-EEG DL networks will be considered:

• EEGNet [36]: The process begins with a temporal convolution, which is followed by a depthwise layer that serves as a spatial representation for each filter produced at the previous stage. Afterward, an exponential linear activation function (ELU) is used before an average pooling and a dropout to help minimize overfitting. Next, it applies a separable convolution, which is followed by another ELU activation and an average pooling. Lastly, a second dropout layer precedes the flattening and classification stages. Batch normalization is always performed immediately after each convolutional layer.
• KREEGNet [38]: Similar to EEGNet, it uses a Gaussian kernel after batch normalization to extract the connectivity between EEG channels. A delta kernel is implemented on the label data, and a Centered Kernel Alignment (CKA)-based regularization between connectivities and label data is added as a penalty to the straightforward cross-entropy.
• KCS-FCNet [39]: A single convolutional stage before using a gaussian kernel to measure EEG connectivity is utilized. These are then run through an average pooling layer before batch normalization and classification. Interestingly enough, a dropout step is done between the flatten layer and the dense layer.
• ShallowConvNet [61]: It performs two consecutive convolutions, then proceeds with batch normalization and square activation. After that, average pooling is done before logarithmic activation and dropout. Finally, a layer of flattening and density is applied for classification.
• DeepConvNet [19]: The system employs two convolutional layers sequentially, followed by batch normalization and ELU activation. Then, a max pooling and a dropout are employed before another convolutional layer and batch normalization. Another set of ELU activation, max pooling, and dropouts is performed before a final convolution and batch normalization. Finally, another ELU, max pooling, and dropout are performed before classifying.
• TCFusionNet [37]: Similar to EEGNet, it employs a sequence of residual blocks to gather extra data prior to classification. Each residual block is comprised of a dilated convolution followed by batch normalization, ELU, and dropout twice. In parallel, a 1x1 convolution is done and then concatenated to the output of the residual block. Before flattening and joining the flattened features from the separable convolution stage, multiple residual blocks are put in place in a cascading fashion. Finally, a dense layer is used for classification.

Figure 4 shows a summarized visual guide of the different MI-EEG architectures.

3.3. Layer-Wise Class Activation Maps for Explainable MI-EEG Classification

Layer-wise Class Activation Mapping (Layer-CAM) is a powerful technique for DL interpretability that addresses the need to understand how neural networks make decisions [55]. By generating heatmaps that highlight the most relevant regions in an input image, Layer-CAM allows researchers to visualize the specific features contributing to the network’s predictions at various layers. Unlike traditional CAM methods, Layer-CAM captures class-discriminative regions at intermediate layers, providing a more detailed and multilevel perspective on feature importance. Moreover, it supports model refinement by highlighting misinterpreted regions. Here, we implement Layer-CAM within a MI-EEG framework, by modeling each EEG trial as an image that holds C rows and $\tau$ columns. Then, let ${\mathbf{S}}_l^k\left(\mathbf{X}\right)\in {\mathbb{R}}^{C\times \tau }$ be the upsampled MI-EEG CAM of a given input trial $\mathbf{X}$ for the k-th class ($k\in K$) in the l-th layer ($l\in L$), defined as:

$${\mathbf{S}}_l^k\left(\mathbf{X}\right)=\mathrm{ReLU}\left(\zeta \left(\sum _{p=1}^{{\tilde{P}}_l}{\beta }_{lp}^k\odot {\tilde{\mathbf{Z}}}_{lp}^k\right)\right),$$

(6)

where $\zeta (·)$ is an up-sampling function, $\mathrm{ReLU}\left(x\right)=max(0,x),$${\tilde{\mathbf{Z}}}_{lp}^k\in {\mathbb{R}}^{C_l^{\prime }\times {\tau }_l^{\prime }}$ represents the network activation map for the l-th layer regarding the p-th filter and the k-th MI class, and ${\beta }_{lp}^k\in {\mathbb{R}}^{C_l^{\prime }\times {\tau }_l^{\prime }}$ gathers the corresponding CAM weight matrix with respect to the k-th output score. By utilizing the backward class-specific gradients, Layer-CAM computes each ${\beta }_{lp}^k$ as:

$${\beta }_{lp}^k=\mathrm{ReLU}\left({\displaystyle \frac{\partial {\tilde{y}}^k}{\partial {\mathbf{Z}}_{lp}^k}}\right),$$

(7)

being ${\tilde{y}}^k\in {\mathbb{R}}^{+}$ the k-th class score holding a linear activation.

To highlight relevant spatial and temporal EEG inputs while avoiding spurious CAM artifacts, we normalize the EEG Layer-CAM as:

$${\tilde{\mathbf{S}}}_l^k\left(\mathbf{X}\right)=2{\displaystyle \frac{{\mathbf{S}}_l^k\left(\mathbf{X}\right)}{\underset{l^{\prime }\in L;k^{\prime }\in K}{max}{\mathbf{S}}_{l^{\prime }}^{k^{\prime }}\left(\mathbf{X}\right)}}+{\mathbf{1}}_C{\mathbf{1}}_{\tau }^{\top },$$

(8)

where $\mathbf{1}$ is an all-ones column vector of proper size. Afterward, an EEG explanation map ${\tilde{\mathbf{X}}}_l^k\in {\mathbb{R}}^{C\times \tau }$ can be computed from trial $\mathbf{X}$ as follows:

$${\tilde{\mathbf{X}}}_l^k=\mathbf{X}\odot {\tilde{\mathbf{S}}}_l^k\left(\mathbf{X}\right).$$

(9)

Lastly, the Gain measure (see Equation 10) is used to assess the importance of different regions in an input EEG that contribute to the model’s decision for a specific class [62]:

$$\mathrm{Gain}\left(\mathbf{X}\right|l,k,\widehat{\theta })={\displaystyle \frac{\tilde{f}\left({\tilde{\mathbf{X}}}_l^k\right|\widehat{\theta })-\tilde{f}\left(\mathbf{X}\right|\widehat{\theta })}{|\tilde{f}\left(\mathbf{X}\right|\widehat{\theta })|}},\forall k\in \{-1,+1\}.$$

(10)

3.4. Questionnaire-MI Performance Canonical Correlation Analysis (QMIP-CCA)

Let us consider the questionnaire matrix $\Xi \in {[0,1]}^{R\times Q}$ holding Q informative MI-EEG questions (see Section 3.1) along R subjects. In turn, let $\Gamma \in {\mathbb{R}}^{R\times A}$ be an MI-EEG classification performance matrix, holding the following measures:

$$\begin{array}{cc}\hfill \mathrm{Accuracy}=& {\displaystyle \frac{TP+TN}{TP+TN+FP+FN}}\hfill \end{array}$$

(11)

$$\begin{array}{cc}\hfill \mathrm{AUC}=& {\int }_0^1{\displaystyle \frac{\mathrm{TP}}{\mathrm{TP}+\mathrm{FN}}}d\left({\displaystyle \frac{\mathrm{FP}}{\mathrm{FP}+\mathrm{TN}}}\right)\hfill \end{array}$$

(12)

$$\begin{array}{cc}\hfill \mathrm{Kappa}=& {\displaystyle \frac{2·(TP·TN-FP·FN)}{(TP+FP)·(FP+TN)+(TP+FN)·(FN+TN)}},\hfill \end{array}$$

(13)

where $TP$, $TN$, $FP$, and $FN$ stand for true positive, true negative, false positive, and false negative, respectively. Of note, the Accuracy, Area Under the Curve (AUC), and the Kappa (see Equations 11-13) quantify the DL MI-EEG classification performance [60]. We also compute the aforementioned MI performance measures using, as input, the explanation maps in Equation 9. Consequently, $A=6$ for a given DL model.

Next, we propose to compute the questionnaire and MI-performance inter-subject matching matrices, $\tilde{\Xi }\in {[0,1]}^{R\times Q}$ and $\tilde{\Gamma }\in {[0,1]}^{R\times A}$, as:

$$\begin{array}{c}\hfill {\tilde{\Xi }}_{rq}={\displaystyle \frac{1}{R}}\sum _{r^{\prime }=1}^R\kappa ({\xi }_{rq}-{\xi }_{r^{\prime }q}|{\sigma }_q^2)={\displaystyle \frac{1}{R}}\sum _{r^{\prime }=1}^{R}exp\left({\displaystyle \frac{-\parallel {\xi }_{rq}-{\xi }_{r^{\prime }q}{\parallel }_2^2}{2{\sigma }_q^2}}\right)\end{array}$$

(14)

$$\begin{array}{c}\hfill {\tilde{\Gamma }}_{ra}={\displaystyle \frac{1}{R}}\sum _{r^{\prime }=1}^R\kappa ({\gamma }_{ra}-{\gamma }_{r^{\prime }a}|{\sigma }_a^2)={\displaystyle \frac{1}{R}}\sum _{r^{\prime }=1}^{R}exp\left({\displaystyle \frac{-\parallel {\gamma }_{ra}-{\gamma }_{r^{\prime }a}{\parallel }_2^2}{2{\sigma }_a^2}}\right),\end{array}$$

(15)

with ${\xi }_{rq}\in \Xi$ and ${\gamma }_{ra}\in \Gamma$; $r,r^{\prime }\in \{1,2,\cdots ,R\}$, $q\in \{1,2,\cdots ,Q\},$ and $a\in \{1,2,\cdots ,A\}$. The kernel function $\kappa (·,{\sigma }^2)$ is set as a Gaussian similarity with bandwidth ${\sigma }^2\in {\mathbb{R}}^{+}$, which is found by taking the median of the Euclidean distances between the kernel input samples as: ${\sigma }_q^2=\mathrm{Median}\{\parallel {\xi }_{rq}-{\xi }_{r^{\prime }q}{\parallel }_2:\forall r,r^{\prime }\in R\}$ and ${\sigma }_a^2=\mathrm{Median}\{\parallel {\gamma }_{ra}-{\gamma }_{r^{\prime }a}{\parallel }_2:\forall r,r^{\prime }\in R\}.$

Lastly, to code the questionnaire and MI-performance non-linear matching between subjects by projecting them into a higher-dimensional feature space using $\kappa (·,{\sigma }^2)$, we employ the following CCA-based optimization [63]:

$$\begin{array}{cc}\hfill {\widehat{\alpha }}_m^{\Xi },{\widehat{\alpha }}_m^{\Gamma }=& arg\underset{{\alpha }_m^{\Xi },{\alpha }_m^{\Gamma }}{max}\sum _{m=1}^M\langle \tilde{\Xi }{\alpha }_m^{\Xi },\tilde{\Gamma }{\alpha }_m^{\Gamma }\rangle \hfill \\ \hfill & \mathrm{s}.\mathrm{t}.\parallel \tilde{\Xi }{\alpha }_m^{\Xi }{\parallel }_2=1\hfill \\ \hfill & \parallel \tilde{\Gamma }{\alpha }_m^{\Gamma }{\parallel }_2=1,\forall m\in M;M\le min(A,Q).\hfill \end{array}$$

(16)

The constraint optimization problem in Equation 16, with ${\alpha }^{\Gamma }\in {\mathbb{R}}^A$ and ${\alpha }^{\Xi }\in {\mathbb{R}}^Q$, can be solved by the following generalized eigenvalue-based solution:

$$\begin{array}{cc}\hfill {\left({\tilde{\Gamma }}^{\top }\tilde{\Gamma }\right)}^{-1}{\tilde{\Xi }}^{\top }\tilde{\Gamma }{\left({\tilde{\Xi }}^{\top }\tilde{\Xi }\right)}^{-1}{\tilde{\Xi }}^{\top }\tilde{\Gamma }{\tilde{\alpha }}_m^{\Gamma }& ={\nu }_m{\alpha }_m^{\Gamma },\hfill \end{array}$$

(17)

$$\begin{array}{cc}\hfill {\displaystyle \frac{{\left({\tilde{\Xi }}^{\top }\tilde{\Xi }\right)}^{-1}}{\sqrt{{\nu }_m}}}{\tilde{\Xi }}^{\top }\tilde{\Gamma }{\tilde{\alpha }}_m^{\Gamma }& ={\tilde{\alpha }}_m^{\Xi };{\nu }_m\in \mathbb{R}.\hfill \end{array}$$

(18)

The weights in ${\tilde{\alpha }}_m^{\Xi }$ and ${\tilde{\alpha }}_m^{\Gamma }$ code the relevance of the Q questions and the A MI performance measures to matching the multimodal feature spaces. Besides, the basis matrices: ${\Lambda }^{\Xi }=\left[{\tilde{\alpha }}_1^{\Xi },{\tilde{\alpha }}_2^{\Xi },\cdots ,{\tilde{\alpha }}_M^{\Xi }\right]\in {\mathbb{R}}^{Q\times M}$ and ${\Lambda }^{\Gamma }=\left[{\tilde{\alpha }}_1^{\Gamma },{\tilde{\alpha }}_2^{\Gamma },\cdots ,{\tilde{\alpha }}_M^{\Gamma }\right]\in {\mathbb{R}}^{A\times M},$ allow us to compute the subspaces ${\Psi }^{\Xi }=\tilde{\Xi }{\Lambda }^{\Xi }\in {\mathbb{R}}^{R\times M}$ and ${\Psi }^{\Gamma }=\tilde{\Gamma }{\Lambda }^{\Gamma }\in {\mathbb{R}}^{R\times M},$ and the relevance vectors ${\rho }^{\Gamma }\in {[0,1]}^A,{\rho }^{\Xi }\in {[0,1]}^Q$, yielding:

$${\rho }^{\iota }=\mathrm{softmax}\left({\Lambda }^{\iota }{\left({\Psi }^{\iota }\right)}^{\top }{\Psi }^{\iota }{\mathbf{1}}_M\right),\iota \in \{\Xi ,\Gamma \};$$

(19)

being $\mathrm{softmax}(·)$ the softmax function.

3.5. Multimodal and Explainable Deep Learning Implementation Details

Our Multimodal and Explainable Deep Learning (MEDL) pipeline can be summarized as in Figure 5. First, the GiGaScience dataset is used to train a subject-dependent DL classifier based on convolutional networks (see Section 3.2 and Figure 4 for DL architectures). Then, a Layer-CAM-based approach is carried out to visualize and quantify the MI-EEG explainability (see Section 3.3). Lastly, QMIP-CCA approach is used to compute the questionnaire and the MI-EEG performance matching (see Section 3.4).

We trained each model using TensorFlow version 2.17.0 and Keras version 3.2.1. All tests were done in Kaggle notebook environments. These environments provide two Tesla T4 GPUs with 15GB of VRAM, 30GB of RAM, an Intel Xeon CPU @ 2GHz with two threads per core, and two sockets per core. We fixed 500 epochs, terminated it on nan, and reduced the learning rate on plateaus as callbacks. Also, Adam optimizer and the categorical cross-entropy-based loss are fixed. Table 1 summarizes the DL hyperparameters employed. Any values not mentioned are fixed regarding TensorFlow’s default selection. The approaches were trained using stratified cross-validation for five folds, primarily measuring accuracy to determine the best model for each subject. For this research, the best fold was selected for each model and subject to generate the best CAM possible for each subject and network. Additionally, all notebooks and codes are publicly available at https://github.com/Marcos-L/CAMs-Enhancements.

4. Results and Discussion

4.1. MI Classification Performance

All six CNN models were trained, evaluated, and contrasted among themselves based on their cross-validation results. Figure 6 shows the different metrics for each model prior to the CAM-based enhancements. KREEGNet shows the best accuracy and AUC score, while DeepConvNet shows the worst. In general, kernel methods improve upon the baseline EEGNet. Despite having fewer feature extraction layers, ShallowConvNet remains comparable, while DeepConvNet, due to its higher parameter count, loses a significant amount of accuracy compared to the other techniques. Moreover, despite its more complex architecture, TCFusion performs almost identically to EEGNet. This shows that simple and shallow models outperform deeper networks, implying that complex strategies may overfit more easily when training subject-specific MI-EEG tasks.

Furthermore, when observing inter-subject accuracy across models, as seen in Figure 7, DeepConvNet’s low performance becomes more apparent. Grouping subjects into three groups based on their classification accuracy on EEGNet (good, mid, and poor), DeepConvNet struggles a lot to properly classify good MI subjects, only ever outperforming EEGNet for the two worst subjects. In contrast, KCS-FCNet, although underperforming for good subjects, remains consistent for mid- and poor subjects. Interestingly, ShallowConvNet performs well in the poor performing group, indicating that the EEG for these subjects is highly contaminated, making fewer features more effective for classification than more.

Figure 8 shows the distribution of subject performance across groups for each model. It is clear that DeepConvNet yields lower results compared to other models such as EEGNet, which are capable of producing comparable results. I This is likely due to DeepConvNet’s greater parameter count and higher number of layers, which cause a slight overfitting problem and result in less refined early feature extraction than could be expected.

4.2. Explainable MI-EEG Classification Results

Figure 9 displays the average class score percentage gain for EEGNet across performance groups and class labels. This is used to look at how the proposed solution affected the class score. Figure 9a,b show that the good and mid-performing groups are biased towards a specific class, with only the poor performing subjects presenting improvements for both. In contrast, ShallowConvNet’s percentage gains presented in Figure 10 consistently have all groups improve for both classes. On the other hand, TCFusion barely shows any gain at all. Figure 11 shows both good and mid-performing groups only having tiny amounts of improvement when compared to the other two models. The only group to have substantial improvements is the poor performing one, but only for right-hand MI.

When examining the CAMs used per class, it is possible to observe a pattern emerge. Figure 12a,c show the average topomaps for good and poor performing subjects. The good-performing subject has the most important features located in the right sensorimotor area, with no information being highlighted on the left side. Then, the model learns the features related to one class and classifies the other whenever the information in this area is non-conclusive, which explains the bias presented by the percentage gain. In contrast, the poor-performing subject has information highlighted all across the head, with both classes highlighting almost the same regions. However, subjects with average performance, like the one shown in Figure 12b, appear to be a mix of the two. The most important information is found in the sensorimotor area for one side, while the rest of the EEG is messed up by noise. This behavior is consistent between models, except for TCFusion, as presented in Figure 15, which generates spatially noisy CAMs, possibly explaining why the boost-to-class score is minimal when CAM enhancement is applied.

Figure 12. Topomaps of EEGNet. The top row shows the maps for the left-hand class while the bottom row shows the same for the right-hand class. Topomaps are min-max normalized horizontally.

Figure 12. Topomaps of EEGNet. The top row shows the maps for the left-hand class while the bottom row shows the same for the right-hand class. Topomaps are min-max normalized horizontally.

Figure 13. Topomaps of KREEGNet. The top row shows the maps for the left-hand class while the bottom row shows the same for the right-hand class. Topomaps are min-max normalized horizontally.

Figure 13. Topomaps of KREEGNet. The top row shows the maps for the left-hand class while the bottom row shows the same for the right-hand class. Topomaps are min-max normalized horizontally.

Figure 14. Topomaps of ShallowConvNet. The top row shows the maps for the left-hand class while the bottom row shows the same for the right-hand class. Topomaps are min-max normalized horizontally.

Figure 14. Topomaps of ShallowConvNet. The top row shows the maps for the left-hand class while the bottom row shows the same for the right-hand class. Topomaps are min-max normalized horizontally.

Figure 15. Topomaps of TCFusion. The top row shows the maps for the left-hand class while the bottom row shows the same for the right-hand class. Topomaps are min-max normalized horizontally.

Figure 15. Topomaps of TCFusion. The top row shows the maps for the left-hand class while the bottom row shows the same for the right-hand class. Topomaps are min-max normalized horizontally.

Finally, Figure 16 shows the improvement distribution for the different models across all three groups. Noticeably, ShallowConvNet has consistently positive improvements as oppose to other models which have a mix of improvements and losses. However, in general, all models tend to show improvements or at least remain the same with a median improvement close to zero across all three groups, with some edge cases showing massive losses. The only exception to this rule is TCFusion, which consistently loses accuracy across all three groups. This then implies that model improvement is tied to the complexity of the neural network, as shallow architectures either improve or remain consistent.

4.3. Questionnaire and MI-EEG Performance Relevance Analysis Results

Figure 17 shows the CCA and our kernel-based CCA variant relevance analysis for the questionnaire data compared to the MI-EEG performance measures. While linear CCA only yields a single correlation for the "How do you feel?" question, which spans from "very good" to "very bad" or "tired," our approach reveals a more diverse range of questions. The questions that have a relevance value greater than or equal to 0.5 pertain to the patient’s overall feelings across various runs, with the exception of a single question with an expected relevance of 0.54. On the other hand, CCA says that DeepConvNet’s kappa after improvements is the most important MI performance. On the other hand, kernel-based CCA says that EEGNet’s accuracy after improvements and kappa after improvements are the most important features. These results suggest how a person feels during the experiment relates to the expected improvements produced by the feedback. As these improvements come from the CAMs, the results imply a connection between the regions observed by the model and how comfortable or relaxed the subject feels. It is also likely that EEGNet is the most important model as it is the basis for the other CNNs, with KREEGNet being the second most important as it is a direct evolution.

5. Conclusions

We present a Multimodal and Explainable Deep Learning (MEDL) framework for MI-EEG classification, combining Class Activation Maps (CAMs) and Canonical Correlation Analysis (CCA) to improve classification accuracy and enhance the interpretability of MI-EEG-based models. The proposed approach involves evaluating various deep learning approaches for MI-EEG classification. Additionally, the use of CAM-based methods successfully highlighted relevant patterns crucial for decision-making in motor imagery classification. Using the Questionnaire-MI Performance Canonical Correlation Analysis (QMIP-CCA) framework gave us a new way to connect subjective questionnaire data with MI-EEG classification performance. This showed us important physiological and cognitive factors that affect how well the models explain things and how accurate the classifications are. ShallowConvNet, in particular, showed the most consistent improvements across different performance groups, proving to be more robust in handling noisy EEG data for poorly performing subjects.

For future work, we aim to extend the multimodal approach by integrating more diverse physiological and environmental data sources to further enhance model accuracy and interpretability. Additionally, exploring more advanced explainability methods, such as Transformer-based networks and more sophisticated attention mechanisms, could improve the robustness of CAMs in capturing relevant EEG features across a wider variety of tasks. Another direction would be the development of subject-independent models to address inter-subject variability, which remains a significant challenge in EEG classification.