RAM-W600: A Multi-Task Wrist Dataset and Benchmark for Rheumatoid Arthritis

TL;DR

RAM-W600 provides the first public, multi-task wrist radiograph dataset for rheumatoid arthritis, enabling both wrist bone instance segmentation and SvdH bone erosion scoring. It includes 1048 PA wrist images from 388 patients across six centers, with 618 segmentation masks and 4800 BE scores, supporting tasks such as JSN progression quantification and erosion detection. Benchmarking across a wide range of architectures reveals strong overall segmentation performance but persistent boundary and erosion-related challenges, and BE classification remains difficult due to severe class imbalance. This dataset offers a standardized, high-quality resource to advance CAD-based wrist RA analysis and longitudinal disease monitoring in diverse clinical settings.

Abstract

Rheumatoid arthritis (RA) is a common autoimmune disease that has been the focus of research in computer-aided diagnosis (CAD) and disease monitoring. In clinical settings, conventional radiography (CR) is widely used for the screening and evaluation of RA due to its low cost and accessibility. The wrist is a critical region for the diagnosis of RA. However, CAD research in this area remains limited, primarily due to the challenges in acquiring high-quality instance-level annotations. (i) The wrist comprises numerous small bones with narrow joint spaces, complex structures, and frequent overlaps, requiring detailed anatomical knowledge for accurate annotation. (ii) Disease progression in RA often leads to osteophyte, bone erosion (BE), and even bony ankylosis, which alter bone morphology and increase annotation difficulty, necessitating expertise in rheumatology. This work presents a multi-task dataset for wrist bone in CR, including two tasks: (i) wrist bone instance segmentation and (ii) Sharp/van der Heijde (SvdH) BE scoring, which is the first public resource for wrist bone instance segmentation. This dataset comprises 1048 wrist conventional radiographs of 388 patients from six medical centers, with pixel-level instance segmentation annotations for 618 images and SvdH BE scores for 800 images. This dataset can potentially support a wide range of research tasks related to RA, including joint space narrowing (JSN) progression quantification, BE detection, bone deformity evaluation, and osteophyte detection. It may also be applied to other wrist-related tasks, such as carpal bone fracture localization. We hope this dataset will significantly lower the barrier to research on wrist RA and accelerate progress in CAD research within the RA-related domain.

Figures (11)

• Figure 1: Overview of the RAM-W600 dataset, designed for wrist bone segmentation and SvdH BE scoring tasks. (MC 1 to 5: Metacarpal 1 to 5; Tz: Trapezoid; Tr: Trapezium; Sca: Scaphoid; Radius: DistalRadius; Cap: Capitate; Ham: Hamate; Lu: Lunate; Tri: Pisiform $\&$ Triquetrum; Ulna: DistalUlna)
• Figure 2: Distribution and Statistics for the age, gender, institution, number of shots, and BE scores in the RAM-W600 dataset. (A) Circular overview of the internal cohorts. Each bar around the circular plot represents a unique patient. The concentric layers from inner to outer encode: (i) Gender distribution. (ii) Institution distribution. (iii) SvdH BE scores in both wrists for each study. Patients with multiple studies are represented multiple times in this layer. (iv) The phase of imaging and the patient’s age at the time of each acquisition. (B) Circular overview of the external validation data. Similar to (A), each bar around the circular plot represents a unique patient. (C) Distribution of SvdH BE scores by joint surface.
• Figure 3: Wrist bone segmentation visualization results. The solid box indicates segmentation challenges caused by BE, while the dashed box represents difficulties arising from bone overlap.
• Figure 4: BE $\&$ nonBE confusion matrix results for classification of BE.
• Figure 5: A total of 1048 DICOM-format wrist radiographs were collected, including 916 internal cases from our institutions and 132 external cases from three different sources. Within the internal cohort, 116 images were identified as non-RA, while the remaining were RA cases. All 132 external images were non-RA. After filtering, 430 advanced RA cases with bony ankylosis were excluded. The final dataset was used for two primary tasks: wrist bone instance segmentation (618 BMP images) and BE classification (800 images × 6 joint areas). The external non-RA images were used exclusively for comparison purposes.