Cristian Lazo Quispe

Cristian Lazo Quispe

ML Researcher · Medical Imaging

generative models · anomaly detection · brain MRI · multimodal AI

I am a computer vision researcher focused on medical imaging - generative modeling and unsupervised anomaly detection for brain MRI, and reconstruction-based approaches for neurodegenerative disease analysis. I am currently mentored by Postdoc. Juan Miguel Valverde (DTU Compute) through the RISE-MICCAI program, developing an age-conditioned flow matching framework for continuous brain-age interpolation in MRI (target: MICCAI 2027); I previously explored unsupervised anomaly detection in brain MRI with Dr. Liam Chalcroft (UCL) under the Fatima Fellowship.

Beyond neuroimaging, I have two first-author challenge papers at MICCAI 2025, a research year at Kyoto University, and 5+ years of production ML engineering. I won 1st place at the CVPR 2026 MSLR Challenge. I teach two courses to physicians and clinical managers at Universidad Científica del Sur - clinical Big Data, and machine-learning methods - and probability and statistics for the MIT MicroMasters in Peru, where I have published all 13 sessions of materials. Outside research, I build open AI tools for civic life, health, and accessibility - see my projects.

Research interests

Toward reliable, interpretable generative models for brain MRI - reconstruction- and mask-based anomaly detection that flags pathology without labelled lesions, robust to low-field and out-of-distribution scans.

Denoising diffusion & flow matching · reconstruction-based anomaly detection · segmentation and generalization for clinically deployable neuroimaging.

Training at scale

I train large multimodal models on multi-node, multi-GPU clusters - PyTorch DDP and FSDP under SLURM, with mixed precision, gradient checkpointing, and sharded optimizer state. Runs are instrumented end to end: MLflow and Weights & Biases for tracking, PyTorch Profiler for throughput and memory. The same pipelines carry into production - Docker, Kubernetes, and CI/CD on GCP and AWS.

Reference implementation →

News

  • Jul 2026 Presented the poster Age-Conditional Latent Flow Matching for Modeling Healthy Brain Ageing in MRI at DL4Sci 2026, UC Berkeley.
  • Jul 2026 Selected participant at the DL4Sci Summer School 2026 (Jul 20–24) - a 5-day intensive on foundation models, reasoning, and agentic AI for science at UC Berkeley, run by Berkeley Lab and NERSC with 230+ researchers. Speakers included researchers from DeepMind, Anthropic, NVIDIA, Stanford, and NYU.
  • Jun 2026 1st place at the CVPR 2026 MSLR Challenge (Track 2) with team AI4Good - 0.931 Top-1 accuracy on radar-based multimodal sign-language recognition.
  • Sep 2025 Started the RISE-MICCAI mentorship on generative modeling for brain MRI.
  • 2025 Two challenge papers at MICCAI 2025 (LISA - Springer LNCS; RISE).
  • 2021 1st place, UTEC Diploma in Big Data & ML against COVID-19 - deep-learning classification of COVID-19 severity in chest X-rays.
  • 2020 Co-built OXIPUL, a mechanical ventilator prototype, as software & embedded-systems lead; presented to and endorsed by Peru's Minister of Health at UNI (press: RPP).
  • 2020 Built a real-time masked-face recognition access-control system for COVID-19 safety protocols at MDP Consulting - featured on national TV news.

Selected work

  • Robust Multi-Label Classification of MRI Artifacts in Low-Field Neonatal Brain Imaging
    Cristian Lazo, Roberto Espinoza. MICCAI 2025LISA ChallengeSpringer LNCS · Scopus Q2 papercode
  • AI4Good at SignEval 2026: Radar-Based Isolated Sign-Language Recognition
    Cristian Lazo, Gissella Bejarano. 1st place · MSLRCVPR 2026CVPRW proceedings paper
  • Flow Matching with Contrastive Learning for Continuous Brain-Age Interpolation in MRI
    Cristian Lazo, Juan Miguel Valverde. in progresstarget MICCAI 2027

All publications →