English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Deep learning fostered a leap ahead in automated skin lesion analysis in thelast two years. Those models are expensive to train and difficult toparameterize. Objective: We investigate methodological issues for designing andevaluating deep learning models for skin lesion analysis. We explore 10 choicesfaced by researchers: use of transfer learning, model architecture, traindataset, image resolution, type of data augmentation, input normalization, useof segmentation, duration of training, additional use of SVMs, and test dataaugmentation. Methods: We perform two full factorial experiments, for fivedifferent test datasets, resulting in 2560 exhaustive trials in our mainexperiment, and 1280 trials in our assessment of transfer learning. We analyzeboth with multi-way ANOVA. We use the exhaustive trials to simulate sequentialdecisions and ensembles, with and without the use of privileged informationfrom the test set. Results -- main experiment: Amount of train data hasdisproportionate influence, explaining almost half the variation inperformance. Of the other factors, test data augmentation and input resolutionare the most influential. Deeper models, when combined, with extra data, alsohelp. -- transfer experiment: Transfer learning is critical, its absence bringshuge performance penalties. -- simulations: Ensembles of models are the bestoption to provide reliable results with limited resources, without usingprivileged information and sacrificing methodological rigor. Conclusions andSignificance: Advancing research on automated skin lesion analysis requirescurating larger public datasets. Indirect use of privileged information fromthe test set to design the models is a subtle, but frequent methodologicalmistake that leads to overoptimistic results. Ensembles of models are acost-effective alternative to the expensive full-factorial and to the unstablesequential designs.

Data, Depth, and Design: Learning Reliable Models for Skin Lesion Analysis