ZAINAMIRZA
Biomedical Engineer·Researcher·Applied AI
Building biomedical systems, research workflows, and AI tools that solve real clinical problems.
Recognised for an independent DFU healing prognosis project with real clinical impact potential.
I developed a diabetic foot ulcer healing prognosis model using routinely collected clinical data to support the prediction of healing outcomes. I am incredibly grateful that the hard work, purpose, and potential real-world impact of this research were recognised.
This award is especially meaningful to me because the project was not only academically challenging, but personally challenging too. There were many moments where I doubted myself and questioned whether I was capable of doing research at this level. To have completed this work independently and receive first place among group projects makes this recognition even more special.
Receiving the award
Certificate
Video CV
A one-minute introduction to my background, projects, and interest in biomedical engineering.
Projects
ROC-AUC by Model
Diabetic Foot Ulcer Healing Prognosis
Developed a 4-module machine learning pipeline to predict 12-week healing outcomes in diabetic foot ulcers using routine clinical variables. Applied nested cross-validation, MICE imputation, probability calibration, and threshold optimisation across Logistic Regression, KNN, Random Forest, and XGBoost models.
AUC 0.758 · 87 cases · 4-module pipeline
Thermal Image Registration Pipeline
Built an end-to-end pipeline to register RGB images to thermal images and transfer ulcer segmentation masks into thermal space for diabetic foot ulcer assessment. Integrated Meta's SAM2 for wound segmentation, CLAHE for local contrast enhancement, and affine transformation for precise alignment.
407 image pairs · SAM2 segmentation · Production-grade
EMG Signal
93.5% accuracy
EMG Gesture Music Therapy
Designed a complete biomedical signal processing system that translates forearm EMG into musical notes for Parkinson's disease rehabilitation. Applied bandpass and notch DSP filtering, wavelet feature extraction, and SVM classification on 24-channel high-density EMG data, mapping gestures to MIDI notes with real-time audio feedback.
93.5% accuracy · Forearm gesture to MIDI · Parkinson's rehab
CNN Accuracy
Pneumonia Diagnostic Imaging
Built and benchmarked ML pipelines for automated pneumonia classification from chest X-rays. Evaluated Random Forest, XGBoost, and CNN architectures; the CNN achieved near-perfect accuracy, validating deep learning's superiority for medical image classification tasks.
0.98 CNN accuracy · N > 500 chest X-rays
Knee Valgus Angle
N=30 · Rokoko Smartsuit Pro II
Lunge Biomechanics Study
Led a study using the Rokoko Smartsuit Pro II to objectively quantify knee kinematics during lunge exercise in N = 30 participants. Compared natural vs. corrected lunge technique, measuring knee adduction/abduction angles, joint alignment, and trunk inclination to identify ACL injury risk factors undetectable by visual observation.
N=30 participants · Objective vs visual · ACL risk analysis
CAD Schematic
CATIA 3DX · FEA-optimised · 3D printed
Assistive Device for Quadriparesis
Designed a parametric assistive device for quadriparesis patients using CATIA 3DEXPERIENCE. Performed Finite Element Analysis to optimise structural integrity, minimise weight, and select materials. Iteratively refined using simulated user feedback and produced manufacturing documentation.
FEA-optimised · Parametric design · User-centred
Writing Polish
Browser-based writing assistant with real-time grammar checking, tone refinement, and dialect options (Australian, American, British, Canadian English). Professional, Concise, and Academic tone modes with inline click-to-apply suggestions.
Real-time grammar · 3 tone modes · 4 dialects
Assignment Tracker
Semester planning and grade management tool with teaching-week grouping, colour-coded assignments by weight, smart start-date recommendations, and an interactive grade studio, all stored locally, no cloud required.
Semester planner · Grade studio · Fully local
ROC-AUC by Model
Diabetic Foot Ulcer Healing Prognosis
Developed a 4-module machine learning pipeline to predict 12-week healing outcomes in diabetic foot ulcers using routine clinical variables. Applied nested cross-validation, MICE imputation, probability calibration, and threshold optimisation across Logistic Regression, KNN, Random Forest, and XGBoost models.
AUC 0.758 · 87 cases · 4-module pipeline
Thermal Image Registration Pipeline
Built an end-to-end pipeline to register RGB images to thermal images and transfer ulcer segmentation masks into thermal space for diabetic foot ulcer assessment. Integrated Meta's SAM2 for wound segmentation, CLAHE for local contrast enhancement, and affine transformation for precise alignment.
407 image pairs · SAM2 segmentation · Production-grade
EMG Signal
93.5% accuracy
EMG Gesture Music Therapy
Designed a complete biomedical signal processing system that translates forearm EMG into musical notes for Parkinson's disease rehabilitation. Applied bandpass and notch DSP filtering, wavelet feature extraction, and SVM classification on 24-channel high-density EMG data, mapping gestures to MIDI notes with real-time audio feedback.
93.5% accuracy · Forearm gesture to MIDI · Parkinson's rehab
CNN Accuracy
Pneumonia Diagnostic Imaging
Built and benchmarked ML pipelines for automated pneumonia classification from chest X-rays. Evaluated Random Forest, XGBoost, and CNN architectures; the CNN achieved near-perfect accuracy, validating deep learning's superiority for medical image classification tasks.
0.98 CNN accuracy · N > 500 chest X-rays
Knee Valgus Angle
N=30 · Rokoko Smartsuit Pro II
Lunge Biomechanics Study
Led a study using the Rokoko Smartsuit Pro II to objectively quantify knee kinematics during lunge exercise in N = 30 participants. Compared natural vs. corrected lunge technique, measuring knee adduction/abduction angles, joint alignment, and trunk inclination to identify ACL injury risk factors undetectable by visual observation.
N=30 participants · Objective vs visual · ACL risk analysis
CAD Schematic
CATIA 3DX · FEA-optimised · 3D printed
Assistive Device for Quadriparesis
Designed a parametric assistive device for quadriparesis patients using CATIA 3DEXPERIENCE. Performed Finite Element Analysis to optimise structural integrity, minimise weight, and select materials. Iteratively refined using simulated user feedback and produced manufacturing documentation.
FEA-optimised · Parametric design · User-centred
Writing Polish
Browser-based writing assistant with real-time grammar checking, tone refinement, and dialect options (Australian, American, British, Canadian English). Professional, Concise, and Academic tone modes with inline click-to-apply suggestions.
Real-time grammar · 3 tone modes · 4 dialects
Assignment Tracker
Semester planning and grade management tool with teaching-week grouping, colour-coded assignments by weight, smart start-date recommendations, and an interactive grade studio, all stored locally, no cloud required.
Semester planner · Grade studio · Fully local
Experience
Bachelor of Engineering (Biomedical) (Honours)
RMIT University
Expected July 2026
WAM 79
International Baccalaureate (IB)
International School of London, Qatar
Secondary School · Graduated 2021
ATAR equivalent 95
Research Intern
RMIT University
Nov 2025 – Mar 2026
Melbourne, VIC
- Built an RGB-thermal registration and wound-segmentation workflow for diabetic foot ulcer imaging, covering preprocessing, alignment, and mask transfer steps for clinician-facing research.
- Diagnosed reliability issues across thermal imaging and clinical data handoff workflows, improving traceability of outputs for internal review and downstream modelling.
- Prepared technical reporting, figures, and manuscript material with collaborators, contributing to a first-authored IEEE EMBC 2026 submission.
Honours Capstone: AI Prognosis for DFU Healing
RMIT University
Aug 2025 – Present
Melbourne, VIC
- Built and validated a prognosis workflow for 87 diabetic foot ulcer cases, applying preprocessing, leakage control, structured feature handling, and repeated nested validation.
- Benchmarked Logistic Regression, Random Forest, XGBoost, and KNN models; best configuration achieved AUC 0.758 and Brier 0.208.
- Produced honours reporting and thesis material, with progress report scored 90/100.
Operations & Administrative Assistant
Moving Better
Mar 2024 – Jun 2025
Melbourne, VIC
- Developed and maintained performance dashboards to support operational decisions across client acquisition and service delivery.
- Supported the launch of a new website and conversion flow, contributing to a 35% increase in customer leads.
- Coordinated administrative processes, reporting, and client communications across a fast-moving health and wellness business.
- Planned and executed marketing initiatives to improve reach, engagement, and follow-up.
Seasonal Specialist
Apple
Aug 2023 – Feb 2024
Sydney, NSW
- Delivered high-volume technical support across hardware, software, setup, and service triage, contributing to a 20% increase in customer satisfaction.
- Built fluency with integrated Apple ecosystem diagnostics and troubleshooting workflows across devices and services.
- Translated technical issues into clear next steps for customers in a fast-paced retail environment.
Student Ambassador
Scape
Jul 2023 – Feb 2024
Sydney, NSW
- Coordinated and delivered 10+ community initiatives, including cultural events, exam support, and resident engagement programs, increasing participation by 20%.
- Implemented feedback-led improvements that contributed to a 12% increase in resident satisfaction scores.
- Provided day-to-day peer support and onboarding guidance for 150+ students living in residence.
Skills
ML/Data
Signal Processing
Computer Vision
CAD & Simulation
Lab
Web
Tools
Research
Healing Prognosis of Diabetic Foot Ulcers Using Routine Clinical Variables
Mirza Z., Sari N.N., Ogrin R., Hanna S., Waller K., Polus B., Ekinci E., Kumar D.K.
RMIT University · Bolton Clarke Research Institute · University of Melbourne · Austin & St Vincent's Hospitals
Dataset
N = 51 de-identified DFU patients
Methods
Logistic Regression, Random Forest, XGBoost, KNN / 5-fold stratified CV
Key Result
ROC-AUC 0.655 · PR-AUC 0.791 · Brier 0.222
Demonstrated that small, low-cost clinical datasets with interpretable models can yield clinically meaningful prognosis without expensive imaging or large-scale data collection.
View Manuscript