WiFi sensing relies heavily on blackbox machine learning (ML) models due to the large feature space and complexity. Despite achieving very high accuracies in complex scenarios, the blackbox nature of these ML-based sensing techniques is commonly criticized. This is in fact a major source of mistrust as these models provide very little explanation supporting their decision, while often handling critical applications (e.g., elderly monitoring). In this paper, we investigate explainable artificial intelligence (XAI) techniques to shed light on the decisions and behaviors of such blackbox models. Specifically, we propose eXSense, a workflow designed based on state-of-the-art XAI techniques to analyze the behavior of blackbox models both locally and globally. To demonstrate its potential, we conduct an extensive analysis on two case studies from the recent sensing literature. Finally, leveraging the insights obtained from our analysis, we propose and evaluate changes to these models, thus enhancing their efficiency and reliability. This includes reducing the feature space by at least 80% with no/minimal loss ($\leq 1\%$) to the model accuracy.

Intercellular communication is crucial for organ function, with extracellular vesicles (EVs) acting as common messengers for almost all cells. This study proposes a novel EV-mediated intercellular communication system that uses a modulation technique regulated via altered intracellular-cytosolic calcium dynamics regulation. As a case study, intercellular communication within a cardiac muscle is considered with cardiomyocyte cells serving as transceivers. Through molecular communication theory, we propose a comprehensive model addressing EV release kinetics, propagation, degradation, and uptake. A linear time-invariant Poisson channel model is developed and closed-form expressions, verified through particle-based simulations, are derived for EV detection probabilities at the receiver. A closed-form bit error probability is derived and validated through Monte Carlo simulations. By selecting an optimal receiver threshold: 1) bit error rate (BER) in EV-mediated intercellular communication is robust against adverse effects from cardiac disorders, such as myocardial infarction; 2) transmitter design can be optimized by minimizing actuation signal amplitude and pulse width; 3) BER stays stable across heart rates for different distances, demonstrating the robustness of EV-mediated communication. This study enhances our ability to engineer precise and reliable intra-body cardiac communications, which may offer valuable applications for cardiovascular disease treatment, for example, the realization of biological lead-less multi-nodal pacemakers.

Treating brain diseases with therapeutic particles imposes significant challenges as particles are usually too large to traverse the gaps between endothelial cells in the blood-brain barrier (BBB). Focused ultrasound (FUS) for disruption of the BBB has been proposed as a remedy. However, the extent of disruption and the efficiency of the particle delivery to the regions of interest are highly dependent on FUS sonication parameters. This study investigates the effects of not only FUS sonication parameters but also the therapeutic particle admin-istration scheme by exploiting communication-theoretic channel modeling. Specifically, the particle pathways from blood vessels to hallmarked spots in the brain interstitial space are abstracted as a single-input-multiple-output (SIMO) channel. The channel outputs are then examined through the lenses of communication-theoretic measures such as channel gain, transmission efficiency, signal-to-noise ratio, and bit error ratio. The numerical results are displayed utilizing the available clinical data on six patients with brain cancer. The results show that the proposed approach could be exploited in future studies to maximize the efficacy of the treatment and minimize adverse effects.

Realistic models of the components and processes are required for molecular communication (MC) systems. In this paper, a spheroidal receiver structure is proposed for MC that is inspired by the 3D cell cultures known as spheroids being widely used in organ-on-chip systems. A simple diffusive MC system is considered where the spheroidal receiver and a point source transmitter are in an unbounded fluid environment. The spheroidal receiver is modeled as a porous medium for diffusive signaling molecules, then its boundary conditions and effective diffusion coefficient are characterized. It is revealed that the spheroid amplifies the diffusion signal, but also disperses the signal which reduces the information communication rate. Furthermore, we analytically formulate and derive the concentration Green's function inside and outside the spheroid in terms of infinite series-forms that are confirmed by a particle-based simulator (PBS).

Channel modeling is a challenging vital step towards the development of diffusion-based molecular communication networks (DMCNs). Analytical approaches for diffusion channel modeling are limited to simple and specific geometries and boundary conditions. Also, simulation- and experiment-driven methods are very time-consuming and computationally complex. In this paper, the channel model for DMCN employing the fundamental concentration Green's function (CGF) is characterized. A general homogeneous boundary condition framework is considered that includes any linear reaction systems at the boundaries in the environment. To obtain the CGF for a general DMCN including multiple transmitters, receivers, and other objects with arbitrary geometries and boundary conditions, a semi-analytical method (SAM) is proposed. The CGF linear integral equation (CLIE) is analytically derived. By employing the numerical method of moments, the problem of CGF derivation from CLIE is transformed into an inverse matrix problem. Moreover, a sequential SAM is proposed that converts the inversion problem of a large matrix into multiple smaller matrices reducing the computational complexity. Particle-based simulator confirms the results obtained from the proposed SAM. The convergence and run time for the proposed method are examined. Further, the error probability of a simple diffusion-based molecular communication system is analyzed and examined using the proposed method.

Diffusive molecular communication (DMC) is envisioned as a promising approach to help realize healthcare applications within bounded biological environments. In this paper, a DMC system within a biological spherical environment (BSE) is considered, inspired by bounded biological sphere-like structures throughout the body. As a biological environment, it is assumed that the inner surface of the sphere's boundary is fully covered by biological receptors that may irreversibly react with hitting molecules. Moreover, information molecules diffusing in the sphere may undergo a degradation reaction and be transformed to another molecule type. Concentration Green's function (CGF) of diffusion inside this environment is analytically obtained in terms of a convergent infinite series. By employing the obtained CGF, the information channel between transmitter and transparent receiver of DMC in this environment is characterized. Interestingly, it is revealed that the information channel is reciprocal, i.e., interchanging the position of receiver and transmitter does not change the information channel. Results indicate that the conventional simplifying assumption that the environment is unbounded may lead to an inaccurate characterization in such biological environments.

Through-the-wall imaging for a wall with corrugated surfaces is introduced. For this purpose, linear sampling as a fast qualitative method is exploited. The method uses the background Green's function that is derived based on the extended boundary condition method (EBCM) and Floquet theory. The method is very efficient in handling walls with shallow corrugated surfaces. For deeply corrugated walls, the Green's function is modified due to the ill-conditioned matrices arising in the EBCM. This is addressed by using the truncated singular value decomposition technique. Several examples are presented to quantitatively demonstrate the ability of the method in imaging through the wall with slightly and very rough interfaces.

Two-dimensional reconstruction of concealed objects in a wall with two-scale surface roughness is presented. The linear sampling method (LSM) as a single-frequency qualitative imaging technique is exploited. The method uses the background Green's function along with the multistatic response matrix of the overall structure. The Green's function of the wall with deterministic corrugated interfaces is derived analytically here; hence, the computations are extremely efficient. The transmission and reflection matrices of the boundaries are employed along with the generalized scattering matrix theory to find the Green's function of the geometry with the source in the wall. Since the scattering matrix of each surface is only evaluated once for the whole process, the inversion algorithm becomes very suitable for real-time applications. The method is general and could be extended to any multilayered media as well. The LSM is robust against noise and clutter; therefore, the small-scale random roughness on the interfaces does not prevent the method from detection of the objects, verified by the results.