class SaifulIslamMahi:
"""
Biomedical Engineer | Computational Neuroscience Enthusiast | Medical AI Researcher
Islamic University, Kushtia, Bangladesh ·
"""
def __init__(self):
self.name = "Saiful Islam Mahi"
self.affiliation = "Bio-Imaging Research Lab, Islamic University, Bangladesh"
self.advisor = "Assoc. Prof. Dr. Md. Khairul Islam"
self.breakthroughs = [
"First asymmetric Caputo fractional Wilson–Cowan neural mass model (qE ≠ qI)",
"Novel resting-state fMRI biomarker q̂E = β/2 (r=0.993, N=50 HCP subjects)",
"NMDA–GABAergic double dissociation via resting-state fMRI [first ever]",
]
self.published_in = ["PLOS ONE (2025)"]
self.in_prep = ["bioRxiv manuscript → 2026"]
self.conferences = ["IEEE CS Bangladesh Chapter Summer Symposium 2025 (Presenter)"]
def current_mission(self) -> str:
return (
"Decode the brain's fractional-order dynamics and translate them "
"into clinically actionable neuroimaging biomarkers."
)
def next_destination(self) -> str:
return "PhD in Computational Neuroscience / Medical AI 🎯"Tanjin R, Al-Amin M, Etee JM, Siddika A, Siddiki A, Mahi SI, et al.
"An evaluation of Roluperidone as a promising repurposing candidate for Alzheimer's Disease: A Computational Investigation."
PLOS ONE 20(12): e0338211, 2025. doi:10.1371/journal.pone.0338211
Applied the full computational drug-discovery pipeline — from molecular docking and ADMET profiling to molecular dynamics simulation — to evaluate Roluperidone as an Alzheimer's repurposing candidate.
|
Python · Nibabel · Nilearn · SciPy End-to-end batch pipeline for N=50 HCP subjects across 4 resting-state runs. Extracts CIFTI timeseries (59,412 grayordinates → 200 Schaefer parcels), applies Welch PSD + OLS log-log regression in the 0.01–0.10 Hz BOLD band.
|
Python · NumPy · SciPy · Matplotlib Grünwald–Letnikov Caputo derivative from scratch (M=300, convergence < 0.3%). Computed 141×141 Matignon stability atlas. Identified 3 novel oscillatory regimes including premature Hopf destabilisation (−18.3% shift).
|
|
Python · SciPy · Pandas · Nilearn Merges Control-network q̂E with NIH Toolbox scores. Partial Pearson correlations across 5 cognitive domains (working memory, fluid intelligence, cognitive flexibility, episodic memory, response inhibition) with Benjamini–Hochberg FDR correction.
|
PyTorch · U-Net · MONAI · OpenCV U-Net with transfer learning on the ISIC 2018 dermoscopy benchmark. Class-weighted Dice loss handles foreground–background imbalance.
|
|
TensorFlow · Transformers · BioPython Transformer-GRU-CNN hybrid for error localisation in DNA-based digital storage. Handles substitution, insertion, and deletion noise. Presented at IEEE CS BDC Summer Symposium 2025.
|
PyRx · NAMD · Gaussian · PyMOL Full computational pipeline: target identification → molecular docking → ADMET profiling → molecular dynamics simulation → lead optimisation for Roluperidone as an AD candidate.
|
Languages
Deep Learning & ML
Neuroimaging & Computational Neuroscience
Computational Biology
Scientific Computing
I'm always open to:
🔬 Joint research in computational neuroscience or medical AI
📚 PhD opportunities in neuro-AI, brain imaging, or medical image analysis
🏥 Industry collaborations applying AI to clinical diagnostics
🤝 Open-source projects in neuroimaging or biomedical deep learning
"The brain speaks in fractional time. I'm learning to listen."