SANO: Score-based Anomaly Localization for Dermatology

Alvaro Gonzalez-Jimenez, Simone Lionetti, Ludovic Amruthalingam, Philippe Gottfrois, Marc Pouly, Alexander Navarini

2022

Abstract

Supervised learning for dermatology requires a large volume of annotated images, but collecting clinical data is costly and it is virtually impossible to cover all situations. Unsupervised anomaly localization circumvents this problem by learning the distribution of healthy skin. However, algorithms which use a generative model and localize pathologic regions based on a reconstruction error are not robust to domain shift, which is a problem due to the low level of standardization expected in many dermatologic applications. Our method, SANO, uses score-based diffusion models to produce a log-likelihood gradient map that highlights potentially abnormal areas. A segmentation mask can then be calculated based on deviations from typical values observed during training. We train SANO on a public non-clinical dataset of healthy hand images without ornaments and evaluate it on the task of detecting jewelry within images from the same dataset. We demonstrate that SANO outperforms competing approaches from the literature without introducing the additional complexity of solving a Stochastic Differential Equation (SDE) at inference time”.

Paper Citation

in Harvard Style

Gonzalez-Jimenez A., Lionetti S., Amruthalingam L., Gottfrois P., Pouly M. and Navarini A. (2022). SANO: Score-based Anomaly Localization for Dermatology. In Proceedings of the 1st Workshop on Scarce Data in Artificial Intelligence for Healthcare - Volume 1: SDAIH, ISBN 978-989-758-629-3, SciTePress, pages 53-58. DOI: 10.5220/0011539200003523

in Bibtex Style

@conference{sdaih22,
author={Alvaro Gonzalez-Jimenez and Simone Lionetti and Ludovic Amruthalingam and Philippe Gottfrois and Marc Pouly and Alexander Navarini},
title={SANO: Score-based Anomaly Localization for Dermatology},
booktitle={Proceedings of the 1st Workshop on Scarce Data in Artificial Intelligence for Healthcare - Volume 1: SDAIH,},
year={2022},
pages={53-58},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0011539200003523},
isbn={978-989-758-629-3},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st Workshop on Scarce Data in Artificial Intelligence for Healthcare - Volume 1: SDAIH,
TI - SANO: Score-based Anomaly Localization for Dermatology
SN - 978-989-758-629-3
AU - Gonzalez-Jimenez A.
AU - Lionetti S.
AU - Amruthalingam L.
AU - Gottfrois P.
AU - Pouly M.
AU - Navarini A.
PY - 2022
SP - 53
EP - 58
DO - 10.5220/0011539200003523
PB - SciTePress