Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

MANNERS: A strategy for representation learning in multivariate datasets with high proportions of missing data.

Patterns (New York, N.Y.)·2026
Same author

MRIgRT real-time target tracking: TrackRAD2025 challenge report.

Medical image analysis·2026
Same author

A review of deep learning-based Unsupervised Anomaly Detection in brain MRI.

Medical image analysis·2026
Same author

A-scan sequence transformers for palpation with optical coherence elastography.

Biomedical optics express·2026
Same author

Spatiotemporal remodeling of bone as a reversibly adaptive biological material in Djungarian hamsters under regulated photoperiod conditions.

Acta biomaterialia·2026
Same author

Current perspective in acute ischemic stroke: net water uptake and imaging-guided selection as bridges to personalized, tissue-based care.

Frontiers in neurology·2026

Related Experiment Video

Updated: Dec 24, 2025

Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances
07:35

Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances

Published on: October 11, 2018

7.9K

Skin lesion classification using ensembles of multi-resolution EfficientNets with meta data.

Nils Gessert1,2, Maximilian Nielsen2,3, Mohsin Shaikh2,3

  • 1Institute of Medical Technology, Hamburg University of Technology, Hamburg, Germany.

Methodsx
|April 16, 2020
PubMed
Summary

Our deep learning method won the ISIC 2019 Skin Lesion Classification Challenge by addressing class imbalance and unknown classes. We achieved first place in both tasks using an ensemble of models and incorporating patient metadata.

Keywords:
Convolutional neural networksDeep LearningMulti-class skin lesion classification

More Related Videos

Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
05:56

Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application

Published on: April 14, 2023

3.1K
Supervised Machine Learning for Semi-Quantification of Extracellular DNA in Glomerulonephritis
09:16

Supervised Machine Learning for Semi-Quantification of Extracellular DNA in Glomerulonephritis

Published on: June 18, 2020

7.2K

Related Experiment Videos

Last Updated: Dec 24, 2025

Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances
07:35

Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances

Published on: October 11, 2018

7.9K
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
05:56

Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application

Published on: April 14, 2023

3.1K
Supervised Machine Learning for Semi-Quantification of Extracellular DNA in Glomerulonephritis
09:16

Supervised Machine Learning for Semi-Quantification of Extracellular DNA in Glomerulonephritis

Published on: June 18, 2020

7.2K

Area of Science:

  • Dermatology
  • Computer Science
  • Medical Imaging

Background:

  • Skin lesion classification is crucial for early diagnosis of skin cancer.
  • The ISIC 2019 challenge aimed to advance automated skin lesion classification using dermoscopic images and patient data.
  • Existing methods face challenges like class imbalance, varying image resolutions, and the presence of unknown classes.

Purpose of the Study:

  • To develop a robust deep learning-based method for the ISIC 2019 Skin Lesion Classification Challenge.
  • To achieve state-of-the-art performance in classifying skin lesions from dermoscopic images and patient metadata.
  • To address specific challenges within the dataset, including class imbalance and an unknown class category.

Main Methods:

  • An ensemble of deep learning models (EfficientNets, SENet, ResNeXt WSL) was employed, selected via a search strategy.
  • Multiple input resolutions and two cropping strategies were utilized during training, coupled with multi-crop evaluation.
  • A data-driven approach handled the unknown class, while loss balancing mitigated severe class imbalance.
  • Patient metadata was incorporated using a dense neural network branch.

Main Results:

  • The proposed deep learning method achieved first place in both Task 1 (image-based classification) and Task 2 (image and metadata-based classification) of the ISIC 2019 challenge.
  • The method effectively addressed challenges of class imbalance and the presence of an unknown class, demonstrating superior performance.
  • The integration of patient metadata significantly contributed to the improved classification accuracy.

Conclusions:

  • The developed deep learning ensemble method provides a highly effective solution for skin lesion classification.
  • The strategies employed, including loss balancing and metadata integration, are crucial for handling complex, imbalanced datasets in medical imaging.
  • This approach sets a new benchmark for automated skin lesion analysis and has implications for clinical diagnosis.