dermoscopy

Project Imaging-X: A Survey of 1000+ Open-Access Medical Imaging Datasets for Foundation Model Development

Skin lesion image analysis that draws on meta-learning to improve performance in low data and imbalanced data regimes.

The souce code of MICCAI'23 paper: Combat Long-tails in Medical Classification with Relation-aware Consistency and Virtual Features Compensation

Deep learning pipeline for multi-type ISIC 2018 skin lesion classification with CNNs, preprocessing, augmentation, and training and inference support 🐙.

Prompt-driven diffusion for jointly generating dermoscopic images and lesion masks, with downstream classification and segmentation evaluation.

Skin Lesion Classification Analysis: A Comparative Study

A deep learning pipeline for skin lesion classification using ISIC dataset with multiple deep learning cnn algorithms and advanced preprocessing including multithreaded loading, augmentation, and performance evaluation.

Автоматическое клиническое описание дерматоскопических изображений: признаки → бакетизация → ранжирование → Qwen2.5-7B

Source code for the paper: "Dermoscopic Dark Corner Artifacts Removal: Friend or Foe?"

EfficientNetB0 trained on HAM10000 — 74.15% accuracy across 7 skin disease classes, Grad-CAM explainability, class-weighted loss for imbalanced medical data, Google Colab T4 GPU

Pixel-level skin lesion segmentation using U-Net trained from scratch on HAM10000 — 0.9115 Dice score, 31M parameters, BCE + Dice loss, live demo on HuggingFace Spaces

CNN ensemble for skin lesion classification — BACC 0.846 ± 0.009 · Clinical threshold calibration · ResNet-50 + DenseNet-121 + EfficientNet-B3 · ISIC 2018 Task 3

derm - dermatology/skin iq - intelligence, AI

TÜBİTAK 2209-A: YOLO + ConvNeXt + Swin ensemble ile mobil cilt kanseri tarama sistemi (mel recall 0.87, cihaz üzerinde ONNX inference)

Risk-aware active learning for skin lesion classification: a dual-metric escalation policy that refers a case to a clinician when either model uncertainty or an independent malignancy-risk head crosses a calibrated threshold. 43% fewer unsafe auto-accepts in 12 of 12 configurations. Under peer review.

Reupload and updated code for Efficient and Effective Automated Digital Hair Removal from Dermoscopy Images by collective of authors from 2016. Code has been updated so it runs on windows and CUDA 13.x An example usage of how to use it with python added.

Projects in Data Science - Bsc in Data Science ITU CPH - Group Penguins

A computer vision and machine learning pipeline for automatic skin cancer detection from dermoscopic images. Achieves high accuracy using traditional ML techniques with preprocessing, feature extraction, and ensemble learning.

Benchmark honesto de descriptores de imagen para clasificar lesiones cutaneas dermatoscopicas (ISIC): 390 combinaciones descriptor x algoritmo, control del efecto de lote y validacion externa entre instituciones. AUC 0.741 fuera del archivo de entrenamiento.

Ablation study on extreme medical class imbalance — ResNet18 + DCGAN synthetic augmentation + Grad-CAM explainability on HAM10000 skin lesion dataset (70:1 imbalance ratio).