Ongoing researchSpectraMed
Explainable, signal-based AI for early detection and severity grading of knee osteoarthritis from X-rays.
Instead of learning from raw pixels, SpectraMed turns medical images into signals using the Radon Transform, then applies custom wavelet learning, FFT-based features, PCA and whitening, and Matrix Pencil differential analysis. Ensemble machine-learning models then classify the result, using far less computing and far smaller datasets than CNN- and Transformer-based methods.
Results so far
- Up to 92% test accuracy and 92% F1-score for hierarchical classification
- 90% (KL0 vs KL1) • 85% (KL2 vs KL3–4) • 84% (KL3 vs KL4)
- Works with small training samples, against the tens of thousands that deep learning typically needs
Now working on
Multi-class grading (KL0–KL4), differential disease signature analysis, cross-validation, multi-dataset evaluation and the SpectraMed platform, extending to other medical imaging applications.
Ongoing researchWeaver Lightweight Stream Cipher
Lightweight cryptography for IoT devices and wireless sensor networks.
Weaver combines ECDH key exchange with HKDF session keys, a compact 40-byte dynamic state, an optimized 8-round ARX permutation and an Encrypt-then-MAC mechanism, and avoids memory-hungry S-boxes.
Results so far
- Maximum differential probability reduced from 0.5044 to 0.0074 after eight rounds
- Keystream passes NIST SP 800-22, Dieharder and ENT randomness suites
- Optimized for Arduino, ESP32, STM32, ARM Cortex-M, FPGA and LoRa sensor nodes
Intended for
Industrial IoT, defence systems, healthcare devices, smart agriculture, environmental monitoring and long-range wireless sensor networks.
Ongoing researchSecure Gradient-Based Hybrid Opportunistic Routing (SGHOR)
Secure routing for wireless sensor networks that stays reliable and energy-efficient when nodes fail or come under attack.
SGHOR is a proposed routing protocol for LoRa mesh and ad hoc networks, designed to enhance multi-hop communication, improve packet delivery, and optimize energy efficiency. It combines gradient-based routing, adaptive neighbor discovery, and opportunistic forwarding to support reliable data transmission across compatible LoRa nodes, including text messages, images, audio, and files.
Key features
- Gradient-based routing enables efficient, destination-oriented packet forwarding.
- Opportunistic forwarding improves delivery across weak and unstable wireless links.
- Adaptive neighbor discovery helps maintain connectivity with reduced routing overhead.
- Link-quality assessment, energy-aware forwarding, and security mechanisms strengthen network resilience.
Intended for
Industrial IoT, smart agriculture, environmental monitoring, disaster-response communication, remote infrastructure monitoring, and decentralized LoRa mesh networks.
