Internship • Jun–Sep 2019 • Medical Imaging

Leukemia Detection API

PyTorch • Keras • SeResNext • Pandas/NumPy • Docker Hub

Automated leukemia classification addressing morphological similarity of cells. github.com/saarques/Blood-Cancer-Diagnosis-API →

PyTorchSeResNextKerasDocker
Leukemia — Microscopic → Preprocess → SeResNext → ~90% → Docker
Microscopic → Pandas/NumPy Augment → SeResNext (PyTorch) → ~90% → Docker Hub API — Click to enlarge →
PROBLEM

Manual microscopy is slow and error-prone — healthy vs leukemic cells look alike. Need a reproducible model that handles small, imbalanced medical data.

APPROACH
  • Preprocessing with Pandas/NumPy — normalization, augmentation for small data.
  • SeResNext via PyTorch/Keras — transfer learning + hyperparameter tuning → ~90% accuracy.
  • Containerized as Docker image on Docker Hub for reproducible medical imaging pipeline.
Pipeline: Microscopic image → Preprocess → SeResNext (PyTorch) → Prediction → Docker API.
RESULT

~90% accuracy via tuned SeResNext, Dockerized for reuse.

Built as an API, not a notebook — same mindset I use for production Spring services.

Stack & links
PyTorch, Keras, SeResNext, Pandas, NumPy, Docker
GitHub → • Docker Hub — sarques/bcpmodel (173 pulls) →