The spotlight talks feature international academics and researchers showcasing their work across the breadth of the conference topics. Read on to find out more about them and their speakers.

Artificial Intelligence and Medical Spotlight Talks
(Day 1, 23 September 2026)

Chaired by Dr Anna Bird and Professor Goran Nenadic. 

Dr Anna Bird is the Biosensors Team Lead in IN-CYPHER. Her research focuses on developing new sensing technologies for disease biomarkers.

Professor Goran Nenadic is a Professor of Computer Science at the University of Manchester, and a former fellow of the Alan Turing Institute.

Applied and Clinical Healthcare Cybersecurity Research: An Introduction to the UC San Diego Center for Healthcare Cybersecurity

Author: Dr Jeff Tully

Presented by Dr Jeff Tully, a security researcher specialising in the intersection of medical technology and patient safety.

Abstract: Cyberattacks against healthcare delivery organizations are patient safety events, yet the interventions hospitals rely on to prevent, withstand, and recover from them remain largely unvalidated in clinical environments. The UC San Diego Center for Healthcare Cybersecurity (CHC) was established to close this evidence gap through applied, clinically grounded research. Co-directed by practicing physicians and staffed by a team spanning clinical medicine, computer science and engineering, biomedical engineering, public health, pharmacy, and health economics, CHC operates across UC San Diego Health, the Jacobs School of Engineering, the Qualcomm Institute, and the Skaggs School of Pharmacy. This structure allows a single group to measure harm at the bedside and to design, deploy, and evaluate countermeasures in live clinical settings. We introduce the Center through three representative projects: an eight-month randomized controlled experiment testing the efficacy of enterprise phishing training across more than 19,500 health system employees; Ransomwhere?, which applies passive internet measurement to detect and characterize cyber-driven outages across hospitals at national scale; and CRASHCART, a rapidly deployable platform for restoring critical clinical and business functions after catastrophic downtime, now field tested more than a dozen times. We further describe CHC’s industry membership model, through which medical device manufacturers work directly with faculty and trainees on shared problems in secure-by-design development and workforce education.

A Verifiable Bayesian Model for Statistical Products

Author: Matt Pratola

Presented by Dr Matt Pratola, an academic and founder with interests in tree regression, Bayesian modelling and verifiable statistical products.

Abstract: Statistical products are widely used in many industry settings. A common use-case is a product derived by a trusted authority, such as a government organization or market regulator. The product is then made available for end-users to calculate predictions that result in actions. However, the issue of verifying whether a claimed prediction was actually generated by the model being assumed has largely been ignored in the Statistics field, despite the importance of verification in many applications. We present a Bayesian tree model that generalizes to many statistical modeling settings of importance, rendering generally useful for many statistical products. Combined with ideas from Merkle proofs, the model construction provides an integrated verification technique, adding a provenance checking mechanism to statistical products.

Fair-FedMOE: Group-Fair One-Shot Federated Learning via Prototype-Guided Experts for Medical Imaging Analysis

Authors: Zengxiang Li, Lingzhao Meng, Shuai Guo, Weishan Zhang, Han Yu, Nan Liu, Daniel Shu Wei Ting, Yuru Liu, Shudong Wang, Tao Chen

Presented by Dr Zengxiang (Shawn) Li, a Principal Investigator at the SingHealth Duke-NUS Artificial Intelligence in Medicine Institute (AIMI) and the Singapore Eye Research Institute (SERI), and an Adjunct Associate Professor at the Singapore University of Technology and Design (SUTD).

Abstract: AI diagnostic models achieve high average accuracy but systematically underperform for vulnerable patient populations. Federated Learning (FL) enables privacy-preserving multi-institutional collaboration [2], yet conventional multi-round FL creates repeated cybersecurity exposure. We propose Fair FedMOE, the first One-Shot Federated Learning (OFL) framework achieving group fairness in medical imaging via a Mixture-of Experts (MoE) architecture with Fairness-Aware Expert Routing (FER) and Prototype-Guided Differential Aggregation (PDA). On diabetic retinopathy (DR) detection across three federated hospital datasets, Fair-FedMOE achieves 82.32% RES-AUC at 87.39% AUC with RETFound [11], a retina-specific foundation model, and 89.22% RES-AUC at 92.45% AUC with DINOv3 [16], a general purpose vision model, all in a single communication round, surpassing all 50-round FL baselines while raising worst-subgroup AUC by +10%. 

Improving Resilience Against Cyberattacks in Healthcare (iReACH): A Cyber Criticality Index

Author: Kunal Rajput

Presented by Dr Kunal Rajput, a Clinical Research Fellow PhD candidate at the Institute of Global Health Innovation, Imperial College London, and a General Surgery Specialty Trainee in London, whose research examines the downstream clinical impacts of cyberattacks on health systems.

Abstract: Cyberattacks can propagate through hospital networks via the same digital pathways that support patient care. We mapped dependencies across 130 NHS England acute trusts and developed a Cyber Criticality Index (CCI), combining three dimensions (Propagation Criticality, Structural Bottleneck Criticality, and Bridge Criticality) to identify trusts generating the greatest system-wide disruption when compromised. Four archetypes emerged: CCI-critical nodes, inter-community connectors, intra-community propagators, and peripheral trusts. Criticality concentrated geographically in a central-south corridor spanning the South Midlands and Thames Valley. The CCI offers NHS cyber operations planners an evidence-based framework for prioritising resilience investment. Future work will quantify patient welfare loss from cyber-induced emergency care disruptions and derive machine-learning-based optimization for patient redistribution during large-scale cyberattacks. 

Hardware Spotlight Talks
(Day 2, 24 September 2026)

Chaired by Dr Vivek Mohan and Professor Sebastian Steinhorst.

Dr Vivek Mohan is the Circuits and Systems Team Lead in IN-CYPHER. His research focuses on hardware security, neuromorphic and bioelectronic circuits and systems, and edge-based intelligence.

Professor Sebastian Steinhorst is an Associate Professor of Embedded Systems and Internet of Things in the Department of Computer Engineering of the School of Computation, Information and Technology at Technical University of Munich (TUM) in Germany.

Evidence-Based Auditing for Third-Party Libraries in C and C++ Binaries

Author: Zhengzi Xu

Presented by Dr Zhengzi Xu, a Senior Research Fellow in IN-CYPHER. His research focuses on AI agents for medical device security.

Abstract: Software supply-chain security increasingly depends on knowing which third-party libraries are present inside deployed binaries. Regulatory pressure around secure software development has made this question more urgent for vendors and public-sector buyers [1], [2]. For C/C++ software, library detection remains difficult because dependencies may be statically linked, dynamically linked, copied from source, stripped, forked, or mixed with shared dependencies. Existing tools often compress these cases into a single similarity score, which hides why a candidate library matched and whether that match reflects real reuse. This extended abstract presents BLADE, a Binary Library Analysis and Detection Engine that treats library detection as evidence-based verification rather than score thresholding. BLADE first retrieves a broad set of candidate libraries, then checks each candidate using identity markers, library metadata, match distribution, and cross-candidate comparison. Across 3,403 stripped binaries and 1,016 libraries, BLADE achieves 97.60% precision, 93.74% recall, and 95.63% F1. A proprietary automo- tive update package gives similar results with 93.79% F1. The main lesson is practical: binary SCA tools need to preserve the evidence behind a library claim, especially when that claim may trigger vulnerability triage, license review, or customer-facing compliance work.

QT-PUF: Quantum Tunneling Leakage Based PUF for Implantable IoMT Devices

Authors: Yueqi Ma, Vivek Mohan, Chip Hong Chang

Presented by Mr Yueqi Ma, an Integrated Circuit Design Research Engineer in IN-CYPHER at Imperial Global Singapore.

Abstract: Implantable and wearable Internet of Medical Things (IoMT) devices demand lightweight, ultra-low-power hardware security primitives, yet existing physical unclonable functions (PUFs) based on arbiters, ring oscillators, or SRAM incur excessive dynamic power, calibration overhead, or reliability degradation unsuitable for implantable operation. This work presents QT-PUF, a gate-tunneling-leakage-based PUF that exploits quantum-mechanical gate leakage arising from process-induced variations in standard 65 nm CMOS. A differential readout circuit with a tunable pseudo-resistor current-to-voltage front end converts picoampere-level leakage variations into stable digital responses without external excitation or post-processing stabilization. Simulation of a 64×64 (4,096-cell) PUF array demonstrates near-ideal entropy (0.9999998), an inter-array fractional Hamming distance of 0.5001, and an average bit error rate below 0.000163 across 10– 70°C and 1.0–1.3 V, with an energy cost of 19.21 fJ/bit outperforming most recent CMOS-based PUF implementations and demonstrating suitability for secure, energy-efficient authentication in next-generation implantable IoMT systems.

Retrofitting Post-Quantum End-to-End Security via Auditable Middleware for Legacy Medical Devices 

Authors: Yuan Lu, Chao Jin, Khin Mi Mi Aung 

Presented by Dr Yuan Lu, a Senior Scientist at A*STAR specializing in applied cryptography for real-world systems.

Abstract: Networked legacy medical devices remain deeply embedded in modern healthcare infrastructure, ranging from bedside remote monitoring systems to infusion pumps and imaging equipment. Despite their critical role and operation in networked environments, many of these devices lack fundamental security capabilities, including support for modern cryptographic protocols and mechanisms for authentication and encryption. The urgency of addressing these vulnerabilities is further amplified by the advent of quantum computing, which threatens to undermine classic public-key cryptography. Compounding this challenge, upgrading legacy medical devices themselves to support post-quantum resilience is typically infeasible, creating a fundamental tension between the need for quantum-resistant security and the practical constraints of legacy devices. To address this tension, we propose a post-quantum-secure middleware architecture that realizes system-level end-to-end encryption (E2EE), along with robust and portable end-user credential management combined with strong user authentication for fine-grained access control. Our approach shifts crypto- graphic functionality from constrained endpoints to an auditable middleware layer, which acts as a security gateway mediating all communication between devices and external systems. By decoupling security evolution from device capabilities, our design enables healthcare infrastructures to adopt quantum-resistant protections in a practical and incremental manner.

Understanding and Mitigating Co-Resident Attacks on BLE Medical Devices

Authors: Bowen Hu, Zong Cao, Vivek Mohan, Viktor Schlegel, Anil Anthony Bharath, Chip Hong Chang 

Presented by Dr Bowen Hu, a Research Fellow in IN-CYPHER. His research focuses on addressing security challenges in medical devices and biometric recognition systems.

Abstracts: Bluetooth Low Energy (BLE) is now the default wireless interface for many smartphone-connected medical devices, but Android authenticates BLE access at the phone rather than at the application. This extended abstract reorganizes a study of ResidentBlue, a co-resident attack class in which an untrusted app, mini program, or web page silently reuses the encrypted BLE channel established by a legitimate companion app. Across six commercial medical devices, the study demonstrates private health-data exfiltration, denial of service, replayed command injection that stops a ventilator, and persistent firmware compromise through unsigned over-the-air updates. The work further shows that application-layer encryption is not a complete remedy: one continuous glucose monitor reused the initialization vector of AES-CFB, reducing first-block encryption to a fixed XOR mask, while other devices remained vulnerable to availability and control attacks even without plaintext recovery. The proposed defense is an Android BLE channel authorization mechanism that prompts the user before any secondary app can access an already-connected peripheral, complemented by device-side authentication, freshness, rate limiting, and signed firmware updates.