Image-based Joint-level Detection for Inflammation in Rheumatoid Arthritis from Small and Imbalanced Data
Shun Kato, Yasushi Kondo, Shuntaro Saito, Yoshimitsu Aoki, Mariko Isogawa · 2026
Read Image-based Joint-level Detection for Inflammation in Rheumatoid Arthritis from Small and Imbalanced Data by Shun Kato, Yasushi Kondo, Shuntaro Saito, Yoshimitsu Aoki, Mariko Isogawa (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-14365.txt.
Introduction
Shun Kato and four co-authors tackle a 2026 arXiv medical imaging problem: spotting joint inflammation in rheumatoid arthritis from home photos. Early diagnosis and tight follow-up matter, since ongoing inflammation can permanently damage joints, yet specialist care often takes a long time to reach. RGB hand images captured at home could speed detection.
The task is hard: positive samples are scarce, data are imbalanced, and inflammation is visually subtle. The paper quantifies that difficulty by building a dedicated dataset, and proposes a detection method designed for small, imbalanced RGB data. The goal is a tool that flags joint inflammation early, helping patients reach care before damage becomes irreversible.
Catalog record
| CREATOR | Shun Kato, Yasushi Kondo, Shuntaro Saito, Yoshimitsu Aoki, Mariko Isogawa |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 23 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-14365
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Image-based Joint-level Detection for Inflammation in Rheumatoid Arthritis from Small and Imbalanced Data Shun Kato, Yasushi Kondo, Shuntaro Saito, Yoshimitsu Aoki, Mariko Isogawa Abstract Rheumatoid arthritis (RA) is an autoimmune disease characterized by systemic joint inflammation. Early diagnosis and tight follow-up are essential to the management of RA, as ongoing inflammation can cause irreversible joint damage. The detection of arthritis is important for diagnosis and assessment of disease activity; however, it often takes a long time for patients to receive appropriate specialist care. Therefore, there is a strong need to develop systems that can detect joint inflammation easily using RGB images captured at home. Consequently, we tackle the task of RA inflammation detection from RGB hand images. This task is highly challenging due to general issues in medical imaging, such as the scarcity of positive samples, data imbalance, and the inherent difficulty of the task itself. However, to the best of our knowledge, no existing work has explicitly addressed these challenges in RGB-based RA inflammation detection. This paper quantitatively demonstrates the difficulty of visually detecting inflammation by constructing a dedicated dataset, and we propose a inflammation detection framework with global local encoder that combines self-supervised pretraining on large-scale healthy hand images with imbalance-aware training to detect RA-related joint inflammation from RGB hand images. Our experiments demonstrated that the proposed approach improves F1-score by 0.2 points and Gmean by 0.25 points compared with the baseline model. 1 Introduction Rheumatoid arthritis (RA) is a common, while refractory autoimmune disease that characterized by synovial inflammation that results in systemic joint damage. Such joint damage can cause severe bone destruction and prevent patients from moving their hands freely. Recent advances in the development of new therapeutics, including biologics and Janus kinase inhibitors have made it possible to aim for remission as a primary goal of management of RA kondo-paradigm . However, missing timely access to appropriate medical care can delay the diagnosis of RA or lead to inadequate control of disease activity. Therefore, detecting active synovial inflammation is critically important in RA management. It is an early manifestation of the disease. However, this requires assessment by rheumatologists and the utilisation of specialised equipment including joint ultrasound, if the inflammation is subtle diagnosis-ra . Whereas, the number of rheumatologists is limited and unevenly distributed across regions, RA patients may not receive appropriate assessment of joint inflammation in their family medicine clinic or have to make a long journey to find out specialists. Therefore, there is a great need for development in technologies that can support remote RA diagnosis and management of disease activities though the detecting active joint inflammations. However, traditional methods for diagnosing RA are known to be poorly suited for telemedicine. Accurate diagnosis has typically relied on X-ray X-ray-kantsu ; pitagorasu-X-ray ; hand-xray-transfer or MRI MRI-1 ; MRI-ultrasound imaging, which require specialized equipment and expertise. As a result, these methods are generally limited to patients who can visit medical facilities. Therefore, for RA detection in telemedicine, it is important to use modalities that do not require specialized equipment or expertise (i.e., RGB images) that can be easily captured by the general patients. Therefore, this paper addresses the detection of RA inflammation in individual finger joints using only RGB images captured by conventional cameras, without the need for specialists or specialized equipment. With this method, a patient experiencing joint pain can take a photo of their fingers to detect the exists of the disease. The patient can use the results to decide whether to visit a hospital or as a reference for receiving remote medical care. However, following three technical challenges exist in RA inflammation detection from RGB hand images. • Firstly, as with many other medical imaging tasks, it is difficult to collect hand and finger image samples from affected patients at scale. Consequently, we need to learn discriminative characteristics of diseased hands from small dataset. • One practical way to address the small dataset is to leverage publicly available hand and finger image datasets; however, almost all images are healthy. Therefore, even if such images can be used for training, it remains necessary to handle the resulting class imbalance. • The third factor is the lack of highly reliable datasets. It is known that rheumatologists’ visual inspection does not always agree with the actual presence or absence of inflammation correlation ; therefore, ground-truth labels for inflammation should be assigned based on ultrasound findings rather than visual inspection. However, to the best of our knowledge, datasets annotated in this manner are currently unavailable. Therefore, to address these challenges, we propose a robust deep learning-based approach for joint-level RA inflammation detection. The main contributions of this work are as follows: (1) We propose a novel deep learning-based framework that providing a scalable solution for remote RA monitoring. (2) We leverage a pre-training strategy using a large-scale healthy hand dataset to learn informative and robust hand feature representations, effectively overcoming the inherent challenges of training with a limited number of RA-positive samples. (3) We introduce an imbalance-aware optimization strategy by incorporating Focal Loss to mitigate the severe class imbalance typical of medical datasets, ensuring high sensitivity even in joints with low inflammation prevalence. (4) We construct a high-reliability RGB hand image dataset where inflammation labels are assigned based on ultrasound findings rather than subjective visual inspection. To the best of our knowledge, this is the first study to provide an objective dataset that bridges the gap between external appearance and internal synovial inflammation. 2 Related Works 2.1 Detection for joint RA inflammation Current standard way of detection for finger joint inflammation in RA relies primarily on physical examination by a rheumatologist. Furthermore, as mentioned in the introduction, detecting subtle arthritis often requires MRI MRI-1 or joint ultrasound MRI-ultrasound , necessitating a visit to a specialized medical facility. Regarding the research that related to RA telemedicine using imaging technology, the detection of bone erosions using conventional X-rays has been proposed. However, these approaches primarily evaluate structural changes resulting from past inflammation, rather than the active joint inflammation targeted in our study. In addition, recent studies have explored the utilizing of thermography as a tool for detecting of joint inflammation, referencing MRI synovitis. However, the main problem is the general applicability of thermography devices as a modality of telemedicine. Compared to these other modalities, RGB image-based methods have the potential to contribute to telemedicine. However, unlike X-ray or MRI, RGB images can only capture the external appearance of the body, making diagnosis based solely on these images a highly challenging task. Notably, Phatak et al. pathak pioneered this field by performing joint-level detection of RA inflammation using InceptionResNetv2 inceptionresnetv2 , thereby validating the potential of CNN-based approaches for this task. However, visual signs may not always perfectly reflect active inflammation correlation , establishing a more robust ground truth would be beneficial. 2.2 Machine Learning Methods for Small/Imbalanced Datasets In image classification tasks, class imbalance of dataset is a very common and significant issue. Class Imbalance causes biased classification and overfitting, and leads to a degraderation of the model performance. Major approaches to address class imbalance are class resampling ADASYN ; SMOTE ; BSMOTE ; under-vs-over and cost-sensitive training cost-sesitive ; syuuseiCE ; MFE ; imbalanced-daiic ; focalloss