gabriel Laurente@alexander_laurente284
1
Events attended
1
Submissions made
2+ years of experience
About me
I'm an Electrical & Electronic Engineering student at UNMSM (Lima, Peru) with a strong background in AI, machine learning, and Python development. I combine engineering intuition with practical coding skills to build solutions that actually work — from intelligent data pipelines to medical imaging systems and optimization frameworks. I enjoy problems where the technical challenge is real: understanding the data, choosing the right approach, and delivering something clean and reliable. My research background includes published work accepted at WCCM-ECCOMAS 2026 and AIMed 2026, which means I think carefully before I build — and my hands-on project experience means I ship results, not just prototypes. Whether you need a custom ML model, a data processing workflow, or an AI-powered tool built from scratch, I'm the kind of developer who asks the right questions first and delivers working code at the end.
🤓 Latest Submissions
Breast Cancer Response Prediction — AMD ROCm
We present a hierarchical multimodal deep learning pipeline for predicting pathologic complete response (pCR) to neoadjuvant chemotherapy in breast cancer, validated on 199 patients from the I-SPY 2 clinical trial. The system integrates three complementary architectures: a frozen RadImageNet DenseNet-121 CNN extracts domain-specific DCE-MRI features across 6 contrast phases; a GATv2 Graph Neural Network encodes clinical molecular data (HR, HER2, TNBC status) using biologically-informed edges; and a hierarchical BiLSTM models the temporal treatment trajectory across 4 timepoints (pre- to post-chemotherapy). GroupWise normalization resolves feature scale disparity across CNN, radiomics, and spatial feature groups, enabling stable gradient flow and reducing seed-to-seed AUC variance from ±0.025 to ±0.011. Molecular subgroup analysis reveals performance consistent with clinical biology: TNBC achieves the highest AUC (0.723) due to pronounced contrast enhancement patterns, while HR+/HER2- presents challenges from severe class imbalance (18% pCR rate, only 14 positive cases). The final model achieves AUC 0.771 ± 0.011 and 81% sensitivity, identifying pCR patients who may avoid unnecessary surgery. The entire pipeline runs on AMD ROCm with optimizations for the RX 9070 XT (gfx1201), demonstrating clinical-grade medical AI on AMD hardware.
10 May 2026
