Skip to content
bziamah.tech
DISAL · University of Ghana

Research

Computer vision and deep learning for healthcare, with an emphasis on models that stay robust once they leave the benchmark.

Google Scholar
ORCID Academic CV
01

Publications

2025Journal article

SARS detection in chest CT scan images using the bootstrapped ViT-B/16 model

Justice Kwame Appati, Bless Ziamah, Herbert Ansah Akrofi, Albert Ankomah Dodoo

Iran Journal of Computer Science, 8(2), 537–551 · Springer

A fine-tuned ViT-B/16 reaches 98.83% accuracy on 2,482 chest CT images, outperforming DenseNet201 and VGG19.

Paper
BibTeX
@article{appati2025sars,
  title     = {SARS detection in chest CT scan images using the bootstrapped ViT-B/16 model},
  author    = {Appati, Justice Kwame and Ziamah, Bless and Akrofi, Herbert Ansah and Dodoo, Albert Ankomah},
  journal   = {Iran Journal of Computer Science},
  volume    = {8},
  number    = {2},
  pages     = {537--551},
  year      = {2025},
  publisher = {Springer},
  doi       = {10.1007/s42044-025-00231-1}
}
02

Current work

MSc thesis · In progress

Early sepsis prediction from clinical time series

Physiological signals are re-encoded as Gramian Angular Field images (2D correlation matrices), so convolutional neural networks can learn pairwise temporal relationships across an entire time window, rather than processing the signal step by step.

The work targets better temporal-dependency capture and robustness to non-stationary clinical data.

Fig. 1: Gramian Angular Field. A 1D signal re-encoded as a 2D image a CNN can read. Illustrative.
Method
Gramian Angular Fields · CNNs
Advisor
Prof. Justice Kwame Appati
Lab
DISAL, University of Ghana
03

Affiliation

Jan 2024 – PresentAccra, Ghana

Graduate Research Assistant

Data Intelligence and Swarm Analytics Laboratory (DISAL), Department of Computer Science, University of Ghana · Advisor: Prof. Justice Kwame Appati

  • Master’s thesis research on early sepsis prediction from clinical time series.
  • Published research on SARS detection in chest CT with a bootstrapped ViT-B/16 (Springer, 2025).
  • Present findings at weekly lab meetings and collaborate on ML applications in critical care, finance, agriculture, climate and medical imaging.

PyTorch · TensorFlow · Vision Transformers · ResNet / EfficientNet · Clinical time series

04

Research interests

i.ML for Healthcare

Clinical prediction and decision support from EHR and physiological data: external validation across hospitals, robustness to distribution shift and missing data, and tools clinicians can trust.

ii.Vision Transformers

Advancing ViT architectures for real-world use in medical diagnostics, biometric systems and security-critical visual understanding.

iii.Deep Learning for Production

Architectures that keep their accuracy under real constraints: low latency, resource efficiency and reliability at scale.

iv.Trustworthy AI

Models that stay reliable under adversarial inputs, distribution shift and novel attacks. Robust by design, not through post hoc defenses.

v.Multi-Agent RL

How intelligent agents collaborate, compete and learn in complex environments, applied to automated decision-making.

vi.ML Systems & Infrastructure

Closing the research-to-deployment gap with model serving, monitoring and continuous learning that keep systems performant over time.

Open to PhD positions and research collaborations.

Especially in ML for healthcare, vision transformers and robust, deployable deep learning.

tziamah1@gmail.comAcademic CV (PDF)