ZAFAR AHMAD — CARDIOVASCULAR RESEARCH BRIEF
Prepared 10 October 2026; updated 11 October 2026. Manuscript statuses last checked 9 October 2026.
Portfolio: https://zafar-ahmad-5359.github.io/research.html
Academic CV: https://zafar-ahmad-5359.github.io/downloads/Zafar-Ahmad-Academic-CV.pdf

I am an electrical engineer and cardiovascular signal-processing researcher with an MS from the National University of Sciences and Technology (NUST), conferred on 18 May 2026. I study electrocardiogram (ECG), phonocardiogram or heart-sound (PCG), and photoplethysmogram or optical-pulse (PPG) signals, connecting clinical data collection, deep learning and wearable prototyping.

MS thesis: Cross-Modal Reconstruction of Cardiac Signals: Decoding ECG-PCG Electromechanical Coupling Dynamics. Supervisor: Dr. Nazia Perwaiz; co-supervisor: Dr. Muhammad Salman Khan.

Question: Can timing inferred from heart sounds support murmur characterisation for unseen patients?
My contribution: Developed reconstruction and phase-analysis workflows, designed experiments and patient-grouped evaluation, led the first-author manuscripts and a blind cross-device audit of the reconstruction. The audit covered 338 record pairs from different recording hardware; it did not establish successful transfer to that hardware.
Current finding: A re-evaluation in October 2026 kept each patient’s recordings in one fold and fitted preprocessing on training data only. Expert-reviewed cardiac-phase boundaries carried useful information for murmur characterisation. The tested synthetic-ECG timing route did not establish an advantage over equal-time windows. This separates the value of accurate phase information from the unestablished benefit of automatically reconstructed timing.
Method: Each patient’s recordings stay within one evaluation group; preprocessing is fitted on training folds only; methods are compared on matched patient folds with paired uncertainty. The downstream evaluation used the public paediatric CirCor DigiScope dataset with ten patient-grouped folds and macro-F1 for each task. Paired ECG-PCG reconstruction data, separate-device reconstruction checks and paediatric downstream murmur evaluation have different roles.
Next question: Can reliable automatic timing preserve that phase information across patients and recording systems? An inferred timing reference is not a replacement for measured diagnostic ECG. Earlier thesis and congress analyses are superseded by the October evaluation and are not evidence for synthetic-ECG timing.

Published first-author conference papers:
- ICAI 2023: Automatic Detection of Paediatric Congenital Heart Diseases from Phonocardiogram Signals. Contribution: preprocessing, four-second segmentation, pitch-shift augmentation and a 1D-CNN for binary paediatric CHD (normal/abnormal) detection. https://doi.org/10.1109/ICAI58407.2023.10136668
- EUSIPCO 2024: Phonocardiogram Classification Based on 1D CNN with Pitch-Shifting and Signal Uniformity Techniques. Contribution: band-selective pitch augmentation and an MFCC-based 1D-CNN for a separate five-class valvular/normal heart-sound benchmark. https://doi.org/10.23919/EUSIPCO63174.2024.10715037
Public proceedings PDF: https://eurasip.org/Proceedings/Eusipco/Eusipco2024/pdfs/0001536.pdf

First-author journal manuscripts:
- PCG-to-ECG Signal Reconstruction Using Conditional GAN with Multi-Scale Attention. IEEE Journal of Biomedical and Health Informatics: minor revision; revised manuscript resubmitted July 2026.
- Synthetic-ECG-gated phonocardiography for paediatric heart-murmur characterisation: a cross-modal deep-learning method. European Heart Journal – Digital Health: major revision requested September 2026.
Neither manuscript is accepted or published.
Second-author manuscript: Parallel multi-scale analysis for paediatric heart disease classification from phonocardiograms (Muhammad Salman Khan, Zafar Ahmad, Hinna Nayab). IEEE Journal of Selected Topics in Signal Processing: submitted November 2025; last reported as under review (status as of 9 October 2026); not accepted or published.

ESC Congress 2026, Munich:
- 28 August: first author and presenter, abstract 83190, “Synthetic ECG-guided temporal gating improves paediatric murmur grading, timing, and shape classification on the CirCor DigiScope dataset”. The later patient-grouped evaluation did not support this automatic-timing benefit (see Current finding above): https://esc365.escardio.org/presentation/326239
- 31 August: co-author and presenter of the concept-bottleneck ECG abstract (abstract 86479). My role: technical supervision and congress presentation: https://esc365.escardio.org/presentation/326198

Current roles: Research Assistant and Team Supervisor on CAREON (Continuous AI-Remote E-Health Operations Network), a NUST wearable research programme, since December 2025, supervising three apprentices; remote Research Assistant in cardiac signal modelling with Qatar University since May 2023. Clinical acquisition experience includes hospital/school recording, collection procedures, consent documents, clinical questionnaires and cardiologist-supported echocardiography-confirmed annotation.
Wearable research: CAREON’s wearable platform is a research prototype (ESP32, ECG and optical pulse sensors, IMU) with pilot recordings from consenting volunteers. Next test: timing fidelity and motion sensitivity before interpreting ECG-PPG features. The ECG-to-peripheral-pulse delay includes pre-ejection time and is not isolated pulse transit time.
Probabilistic coupling (in development, since September 2024): copula-based coupled hidden Markov models relating ECG events to heart-sound phases within each beat; exploratory dependence analysis so far.

Doctoral direction: I bring cardiac acquisition, signal-model development and patient-grouped evaluation experience. My modelling so far is statistical and data-driven; I seek training in physiological modelling and prospective validation, asking whether signal-derived cardiac timing can constrain patient-specific physiological models, and how measurement uncertainty and recording-device differences affect identifiability.

Contact: ahmadzafar577@gmail.com
GitHub profile: https://github.com/ZAFAR-AHMAD-5359
Scholar: https://scholar.google.com/citations?user=D5W9TVwAAAAJ&hl=en