PASSION-derm-ICD-Captions: ICD-10, ICD-11, and Dual-Caption Annotations for the PASSION Dermatology Dataset

PASSION-ICD-Captions: ICD-10, ICD-11, and Dual-Caption Annotations for the PASSION Dermatology Dataset Description This dataset provides an annotation extension to the PASSION dermatology dataset, enriching the original image collection with standardized diagnostic codes and textual descriptions. The original PASSION dataset and project are described in: Gottfrois et al., “PASSION for Dermatology: Bridging the Diversity Gap with Pigmented Skin Images from Sub-Saharan Africa,” MICCAI 2024. Original paper: https://doi.org/10.1007/978-3-031-72384-1_66PASSION project and dataset page: https://passionderm.github.io/ For each image, this extension includes ICD-10 and ICD-11 codes, where applicable, together with two human annotated textual captions per image. These additional annotations are intended to support research at the intersection of dermatology, medical image analysis, clinical coding, and multimodal artificial intelligence. The resource enables researchers to link dermatological images with internationally recognized disease classification systems while also providing natural-language descriptions suitable for tasks such as image–text representation learning, image captioning, multimodal retrieval, classification, and evaluation of vision-language models. This release should be considered an annotation layer for the PASSION dataset rather than a replacement for the original dataset. Image identifiers are provided to enable linkage between the annotations and the corresponding images in PASSION. Users should obtain and use the original images in accordance with the PASSION dataset's access conditions and license. Included annotations ICD-10 codes ICD-11 codes Two captions per image Image identifiers linking the annotations to the corresponding PASSION images The goal of this extension is to increase the clinical and semantic utility of the PASSION dataset and facilitate reproducible research involving dermatological image understanding, standardized disease coding, and multimodal learning.