M. R. A. Jamaludin

Research

Research interests

My research addresses trustworthy medical AI, with a focus on evaluation integrity, data-leakage detection and the honest reporting of results in clinical AI benchmarks. I also work on LLM-based explanation and reasoning layers for clinical decision support, explainable AI in safety-critical domains, and computer vision and multimodal learning for medical data. Hybrid quantum-classical machine learning serves as a methodological area.

Key result: evaluation you can trust

Trustworthy medical AI depends on how a model is evaluated. For hybrid quantum-classical skin lesion classifiers on HAM10000, evaluation protocols, rather than architectures, drive the headline differences reported in the literature.

  1. The audit found a leaking split.

    Partitioning augmented images put variants of the same physical lesion on both sides of the split. An audit found that 1,095 of 4,810 lesions (22.8%) crossed the split boundary, artificially inflating all previously reported metrics.

  2. The correction was verified, not assumed.

    The pipeline was rebuilt around a lesion-level partition with zero image-level and lesion-level overlap, and every experiment was re-run across five random seeds with confidence intervals and paired significance tests on a held-out test set never used for tuning. The same pipeline reached 78.41% ± 0.74 test accuracy under the corrected protocol and 94.57% under the leakage-prone one.

  3. The negative result is reported as one.

    A parameter-matched classical control was statistically indistinguishable from the hybrid model (78.14% versus 78.41%, p = 0.81), so no advantage from the quantum branch is claimed at this scale. It is reported as a controlled negative result rather than a failure.

The contribution is methodological: a reproducible leakage audit, a verified correction protocol, an honest multi-seed baseline and a parameter-matched classical control.

Source: Evaluation integrity in hybrid quantum-classical skin lesion classification: A data-leakage audit and lesion-level correction on HAM10000, IEEE Access 14, 2026. The figures are as stated in the published abstract.

Research projects

Oct 2025 - Aug 2026

Quantum-Classical Hybrid AI for Skin Lesion Classification

  • Built a skin lesion classifier combining quantum circuit-based feature extraction with classical deep learning (weighted feature fusion).
  • Added an LLM-powered explanation layer that generates human-readable clinical interpretations of each prediction.
  • Deployed the system as a web-based diagnostic prototype.
  • Published the evaluation and data-leakage audit in IEEE Access (2026); defended as M.Sc thesis (2026).

Stack: PyTorch, PennyLane, Qiskit, Hugging Face Transformers

Papers: Evaluation integrity in hybrid quantum-classical skin lesion classification: A data-leakage audit and lesion-level correction on HAM10000; Stable training of hybrid quantum-classical skin lesion classifiers via offline data augmentation

Published at IEEE Access

Aug 2024 - Sep 2025

Quantum-Based DNA Sequence Alignment with Noise Handling and Error Detection

  • Developed a noise-resilient quantum alignment algorithm using 3-qubit error-detection encoding, hardware-calibrated noise modelling, and Zero-Noise Extrapolation.
  • Achieved 6/6 correct alignments vs. 1/6 for the baseline (confidence > 0.98), using less than half the qubits of standard QECC.
  • Presented at KSII Spring Conference 2025 (Best Paper Award); extended version published in IEEE Access (2025).

Papers: Enhanced quantum-based DNA sequence alignment with noise handling and error detection; Minimizing qubit requirements in scalable quantum models: Leveraging QECC error detection principles

Published at IEEE Access

Datasets

2022

SIBI sign language alphabet dataset

Images of the Indonesian Sign System (SIBI) alphabet, published on Kaggle for community and academic use. Built for the undergraduate thesis (2022).

Kaggle. Related project: SIBI Sign Language Classifier (Undergraduate Thesis)

Thesis and applied projects

2022

SIBI Sign Language Classifier (Undergraduate Thesis)

  • Developed a CNN-based classifier for Indonesian Sign System (SIBI) alphabet images, reaching 90% accuracy, and deployed it with the Flask framework.
  • Published the dataset on Kaggle for community and academic use.

Dataset: SIBI sign language alphabet dataset on Kaggle

2021, 2022

Augmented Reality Learning Media for Arabic Language Learning

Unity Developer

  • Built AR-based learning media using Unity and C# to support the learning of hijaiyah letters and Arabic object vocabulary, used in classroom and university social-service settings.
  • Contributed to application development, documentation, and user guides.

Usage instructions on GitHub

2021

Rust Detector (Google Bangkit Academy Capstone Project)

Cloud Engineer

  • Architected cloud infrastructure on Google Cloud Platform and integrated a machine learning model with an Android application via a Flask API server.

Usage instructions on GitHub