Skin lesion segmentation using U-Net and good training strategies

TL;DR

This paper improves skin lesion segmentation by applying U-Net with effective training strategies, achieving a Jaccard index of 77.5% on the ISIC 2018 dataset.

Contribution

It introduces specific training strategies that significantly enhance U-Net performance for skin lesion segmentation.

Findings

Achieved 77.5% Jaccard index on ISIC 2018 dataset

Demonstrated the effectiveness of training strategies for CNNs in medical imaging

Improved segmentation accuracy over baseline methods

Abstract

In this paper we approach the problem of skin lesion segmentation using a convolutional neural network based on the U-Net architecture. We present a set of training strategies that had a significant impact on the performance of this model. We evaluated this method on the ISIC Challenge 2018 - Skin Lesion Analysis Towards Melanoma Detection, obtaining threshold Jaccard index of 77.5%.