Skip to content
View mcansumert's full-sized avatar
2 followers · 5 following

Block or report mcansumert

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
mcansumert/README.md

Münevver Cansu Mert

Biomedical engineer working on medical image analysis with PyTorch.

Featured

tuseb-stroke-ct — TUSEB/TEKNOFEST 2021 non-contrast head CT

  • Slice-level detection on held-out session 1: AUC 0.84
  • Small U-Net on held-out session 2 official masks: Dice 0.54
  • Grad-CAM showed gantry shortcut learning; skull-interior masking moved attention onto lesions
  • Not a medical device; Ministry of Health data is not redistributed

Earlier JPEG/EfficientNet-B0 work is a failed baseline (shortcut learning on skull/neck): brain-stroke-detection-efficientnetB0

dermoscopic-cbam-ensemble — 3-class dermoscopy (malignant / benign / pre-malignant)

  • Lesion-level split (no same-lesion leakage): accuracy 0.86, macro F1 0.82, malignant recall 0.81, macro AUC 0.96
  • Image-level 91% is an optimistic split; not the number to cite
  • Two operating points: argmax (high recall) vs precision-constrained threshold
  • Grad-CAM++ on EfficientNet-B0 + ResNet-50 + DenseNet-201 + CBAM
  • Not a medical device; weights and training code are not released; HAM10000/ISIC images are not redistributed

Stack

Python, PyTorch, pydicom, medical CT (HU windowing, segmentation), dermoscopy

Contact

munevvercansumert@gmail.com

Pinned Loading

  1. tuseb-stroke-ct tuseb-stroke-ct Public

    PyTorch TUSEB stroke CT: held-out detection (AUC 0.84) and U-Net Dice 0.54

    Python 2

  2. dermoscopic-cbam-ensemble dermoscopic-cbam-ensemble Public

    3-class dermoscopic lesion ensemble (EfficientNet-B0 + ResNet-50 + DenseNet-201 + CBAM). Lesion-level split, malignant operating points. Research only — not for diagnosis. Weights not released.

    Python 2

  3. Image_Processing_use_filtering Image_Processing_use_filtering Public

    A study was carried out to improve the human skeleton medical image by applying different filtering methods in Python programming language.

    Jupyter Notebook 2