Armin Nikpour’s research while affiliated with The University of Sydney and other places

Introduction Selection of antiseizure medications (ASMs) for newly diagnosed epilepsy remains largely a trial-and-error process. We have developed a machine learning (ML) model using retrospective data collected from five international cohorts that predicts response to different ASMs as the initial treatment for individual adults with new-onset epilepsy. This study aims to prospectively evaluate this model in Australia using a randomised controlled trial design. Methods and analysis At least 234 adult patients with newly diagnosed epilepsy will be recruited from 14 centres in Australia. Patients will be randomised 1:1 to the ML group or usual care group. The ML group will receive the ASM recommended by the model unless it is considered contraindicated by the neurologist. The usual care group will receive the ASM selected by the neurologist alone. Both the patient and neurologists conducting the follow-up will be blinded to the group assignment. Both groups will be followed up for 52 weeks to assess treatment outcomes. Additional information on adverse events, quality of life, mood and use of healthcare services and productivity will be collected using validated questionnaires. Acceptability of the model will also be assessed. The primary outcome will be the proportion of participants who achieve seizure-freedom (defined as no seizures during the 12-month follow-up period) while taking the initially prescribed ASM. Secondary outcomes include time to treatment failure, time to first seizure after randomisation, changes in mood assessment score and quality of life score, direct healthcare costs, and loss of productivity during the treatment period. This trial will provide class I evidence for the effectiveness of a ML model as a decision support tool for neurologists to select the first ASM for adults with newly diagnosed epilepsy. Ethics and dissemination This study is approved by the Alfred Health Human Research Ethics Committee (Project 130/23). Findings will be presented in academic conferences and submitted to peer-reviewed journals for publication. Trial registration number ACTRN12623000209695.

Epilepsy poses a significant global health challenge, driving the need for reliable diagnostic tools like scalp electroencephalogram (EEG), subscalp EEG, and intracranial EEG (iEEG) for accurate seizure detection, localization, and modulation for treating seizures. However, these techniques often rely on feature extraction techniques such as Short Time Fourier Transform (STFT) for efficiency in seizure detection. Drawing inspiration from brain architecture, we investigate biologically plausible algorithms, specifically emphasizing time-domain inputs with low computational overhead. Our novel approach features two hidden layer dendrites with Leaky Integrate-and-Fire (dLIF) spiking neurons, containing fewer than 300K parameters and occupying a mere 1.5 MB of memory. Our proposed network is tested and successfully generalized on four datasets from the USA and Europe, recorded with different front-end electronics. USA datasets are scalp EEG in adults and children, and European datasets are iEEG in adults. All datasets are from patients living with epilepsy. Our model exhibits robust performance across different datasets through rigorous training and validation. We achieved AUROC scores of 81.0% and 91.0% in two datasets. Additionally, we obtained AUPRC and F1 Score metrics of 91.9% and 88.9% for one dataset, respectively. We also conducted out-of-sample generalization by training on adult patient data, and testing on children’s data, achieving an AUROC of 75.1% for epilepsy detection. This highlights its effectiveness across continental datasets with diverse brain modalities, regardless of montage or age specificity. It underscores the importance of embracing system heterogeneity to enhance efficiency, thus eliminating the need for computationally expensive feature engineering techniques like Fast Fourier Transform (FFT) and STFT.

The landscape of artificial intelligence (AI) research is witnessing a transformative shift with the emergence of the Kolmogorov–Arnold network (KAN), presenting a novel architectural paradigm aimed to redefine the structural foundations of AI models, which are based on multilayer perceptron (MLP). Through rigorous experimentation and evaluation, we introduce the KAN–electroencephalogram (EEG) model, a tailored design for efficient seizure detection. Our proposed network is tested and successfully generalized on three different datasets, one from the USA, one from Europe, and one from Oceania, recorded with different front-end hardware. All datasets are scalp EEG in adults and are from patients living with epilepsy. Our empirical findings reveal that while both architectures demonstrate commendable performance in seizure detection, the KAN model exhibits high-level out-of-sample generalization across datasets from diverse geographical regions, underscoring its inherent efficacy and adaptability at the backbone level. Furthermore, we demonstrate the resilience of the KAN architecture to model size reduction and shallow network configurations, highlighting its versatility and efficiency by preventing over-fitting in-sample datasets. This study advances our understanding of innovative neural network architectures and underscores the pioneering potential of KANs in critical domains such as medical diagnostics.

Epilepsy is a chronic neurological condition with various etiologies, and recurrent unprovoked seizures characterize it. Hypersalivation is a recognized symptom of insular-opercular epilepsies. A wide range of symptoms can occur during a seizure, including hypersalivation, autonomic responses, oropharyngeal sensations, visceral sensations, somatosensory disturbances, and emotional manifestations. In this case study, we examine a unique scenario of a patient who experienced predominantly salivary seizures. Hypersalivation, pill-rolling movements, and lip-smacking characterized these seizures. Importantly, the patient became seizure-free after undergoing radiofrequency thermocoagulation (RFTC) with the assistance of Stereoelectroencephalography (SEEG). Our discussion will focus on the treatment approach involving SEEG-guided RFTC and the careful identification of the brain cortex responsible for triggering excessive salivation during seizures.

Epilepsy is a significant global health issue, requiring dependable diagnostic tools like scalp encephalogram (scalp-EEG), sub-scalp EEG, and intracranial EEG (iEEG) for precise seizure detection and treatment. AI has emerged as a powerful tool in this domain, offering the potential for real-time, responsive monitoring. Traditional methods often rely on feature extraction techniques like Short-Time Fourier Transform (STFT), which can increase power consumption, making them less suitable for deployment on edge devices. While large models can improve accuracy without STFT, their size also limits their practicality for edge applications. This study introduces Liquid-Dendrite, a novel bio-inspired model for seizure detection, leveraging Liquid-Time Constant Spiking Neurons (LTC-SN) and dendrites spiking neurons (dSN) with heterogeneous time-constants. The model comprises two hidden layers with dendritic neurons and one layer of liquid-time constant networks. Our model achieves a memory efficacy of 535 KB with 130 K trainable parameters. The model was tested across the most noteworthy epilepsy datasets for scalp EEG (TUH and CHB-MIT) and iEEG (EPILEPSIAE). Our model demonstrated commendable performance, achieving AUROC scores of 83%, 96%, and 93%, respectively, outperforming some existing models in an energy and memory-efficient way. Moreover, we conducted a robustness test by blacking out EEG channels at the inference stage, where we showed the ability of our network to work with fewer channels. We could deploy our tiny model and perform inference at the edge of the Raspberry Pi 5 without the need for additional quantization. This highlights the potential of Neuro-Inspired AI for efficient, small-scale, and energy-embedded AI systems across different brain modalities.

This paper introduces an innovative method for fine-tuning a large multi-label model for abnormality detection, utilizing a smaller trainer and advanced knowledge distillation techniques. It studies the effects of fine-tuning on various abnormalities, noting different improvements based on the Original Model’s performance in specific tasks. The experimental setup, optimized for on-device inference and fine-tuning with limited computational resources, demonstrates moderate yet promising enhancements in model performance post-fine-tuning. Key insights from the study include the significance of aligning the µ-Trainer’s behavior with the Original Model and the influence of hyper-parameters like batch size on fine-tuning outcomes. The research acknowledges limitations such as the limited exploration of loss functions in multi-label models and constraints in architectural design, suggesting potential avenues for future investigation. While the proposed Naive Continual Fine-tuning Process is in its early stages, we highlight this paper’s potential model personalization on long-term data. Moreover, weight transfer in our system is exclusively for fine-tuning; hence, it improves user privacy protection by failing data reconstruction attempts from weights, like an issue with Federated learning models. Our on-device fine-tuning prevents the transferring of data or gradients from the edge of the network to their server. Despite modest performance improvements after fine-tuning, these working layers represent a small fraction (0.7%) of the total weights in the Original Model and 1.6% in the µ-Trainer. This study establishes a foundational framework for advancing personalized model adaptation, on-device inference and fine-tuning while emphasizing the importance of safeguarding data privacy in model development.

Purpose :Wireless power transfer (WPT) for stent-based medical devices in the brain, such as endovascular electrocorticography (endoECoG) devices, faces challenges. Typical stent-based endoECoG consists of electrodes placed on a stent and connected with a set of long lead wires to a subcutaneous chest implant responsible for wireless energy harvesting and data telemetry. Eliminating the long lead wires is not a trivial or straightforward task, introducing a great set of challenges. This work demonstrates a feasible method to deliver power directly to a standard medical stent without modifying its structure, mechanically or electrically. Methods :The proposed system employs a subcutaneous relay that converts inductive coupling to capacitive coupling. This solution not only enhances power transfer efficiency while maintaining minimal invasiveness but also addresses the challenges of unstable contact impedance with capacitive coupling WPT. Experimental validation was performed using real skin, bone, and vessel tissues, and finite element simulations were conducted to confirm model accuracy. Results :Experiments demonstrated over 45 mW of power delivery without exceeding safety limits, sufficient for powering endoECoG devices and biosignal sensors. The system achieved 7.26\% DC-to-DC efficiency, the highest reported for stent-based implants without additional transceivers or specialized stent designs. Results closely matched simulations, confirming practical viability. Safety assessments, including specific absorption rate (SAR) analysis and temperature rise simulations, showed compliance with regulatory standards and minimal risk to surrounding tissues. Conclusion :This work demonstrates a reasonably efficient and safe power delivery to stent-based implants in the brain, considering the anatomical challenges regarding the surgical delivery, paving the way for fully wireless, minimally invasive neuroprosthetic devices. The external device does not require close skin contact, making it suitable for long-term applications and improving patient comfort. Future efforts will optimize system components and address manufacturing challenges to facilitate clinical translation.

Objective. Endovascular brain-computer interfaces (eBCIs) offer a minimally invasive way to connect the brain to external devices, merging neuroscience, engineering, and medical technology. Currently, solutions for endovascular electrocorticography (ECoG) include a stent in the brain with sensing electrodes, a chest implant to accommodate electronic components to provide power and data telemetry, and a long (tens of centimeters) cable travel through vessels with a set of wires in between. Removing this long cable is the key to the clinical viability of eBCIS as it carries risks and limitations, especially for patients with fragile vasculature. Approach. This work introduces a wireless and leadless telemetry and power transfer solution for ECoG. The proposed solution includes an optical telemetry module and a focused ultrasound (FUS) power transfer system. The proposed system can be miniaturised to fit in an endovascular stent, removing the need for long, intrusive cables. Main results. The optical telemetry achieves data transmission speeds of over 2 Mbit/s, capable of supporting 41 ECoG channels at a 2 kHz sampling rate with 24-bit resolution. The FUS power transfer system delivers up to 10 mW of power to the implant through the scalp(6 mm), skull(10 mm), and subdural space(5 mm), adhering to safety limits. Testing on bovine tissue (10 mm thick bone, 7 mm thick skin) confirmed the system’s efficacy. Significance. This leadless and wireless solution eliminates the need for long cables and auxiliary implants, potentially reducing complications and enhancing the clinical applicability of eBCIs. The proposed system represents a step forward in enabling safer and more effective ECoG for a broader range of patients.

Neuromodulation techniques have emerged as promising approaches for treating a wide range of neurological disorders, precisely delivering electrical stimulation to modulate abnormal neuronal activity. While leveraging the unique capabilities of artificial intelligence (AI) holds immense potential for responsive neurostimulation, it appears as an extremely challenging proposition where real-time (low-latency) processing, low power consumption, and heat constraints are limiting factors. The use of sophisticated AI-driven models for personalized neurostimulation depends on the back-telemetry of data to external systems (e.g. cloud-based medical mesosystems and ecosystems). While this can be a solution, integrating continuous learning within implantable neuromodulation devices for several applications, such as seizure prediction in epilepsy, is an open question. We believe neuromorphic architectures hold an outstanding potential to open new avenues for sophisticated on-chip analysis of neural signals and AI-driven personalized treatments. With more than three orders of magnitude reduction in the total data required for data processing and feature extraction, the high power- and memory-efficiency of neuromorphic computing to hardware-firmware co-design can be considered as the solution-in-the-making to resource-constraint implantable neuromodulation systems. This perspective introduces the concept of Neuromorphic Neuromodulation, a new breed of closed-loop responsive feedback system. It highlights its potential to revolutionize implantable brain-machine microsystems for patient-specific treatment.

The landscape of artificial intelligence (AI) research is witnessing a transformative shift with the emergence of the Kolmogorov-Arnold Network (KAN), presenting a novel architectural paradigm aimed to redefine the structural foundations of AI models, which are based on Multilayer Perceptron (MLP). Through rigorous experimentation and meticulous evaluation, we introduce the KAN-EEG model, a tailored design for efficient seizure detection. Our proposed network is tested and successfully generalized on three different datasets, one from the USA, one from Europe, and one from Oceania, recorded with different front-end hardware. All datasets are scalp Electroencephalogram (EEG) in adults and are from patients living with epilepsy. Our empirical findings reveal that while both architectures demonstrate commendable performance in seizure detection, the KAN model exhibits high-level out-of-sample generalization across datasets from diverse geographical regions, underscoring its inherent adaptability and efficacy at the backbone level. Furthermore, we demonstrate the resilience of the KAN architecture to model size reduction and shallow network configurations, highlighting its versatility and efficiency by preventing over-fitting in-sample datasets. This study advances our understanding of innovative neural network architectures and underscores the pioneering potential of KANs in critical domains such as medical diagnostics.