Session Information
Session Type: Poster Session B
Session Time: 10:30AM-12:30PM
Background/Purpose: Sjogren’s Disease (SjD) is characterized by focal sieladenitis in minor salivary gland biopsies (mSGB), which is evaluated using the Focus Score (FS). FS ≥ 1 identification is an integral part of the diagnostic approach and patient classification. However, during mSGB evaluation, FS is not reported in an astounding 17%[1], even among specialized academic centres. Moreover, even between experienced pathologists, some inter-observer variability is introduced[1]. As the need for reliable FS calculation and reporting is illustrated, the application of Artificial Intelligence in mSGB analysis shows promising potential and mandates to be explored.
Methods: A set of consecutive mSGBs was randomly selected from our archive, using only hematoxylin and eosin (H&E) staining. Three of them were excluded from analysis due to quality issues. All biopsies were digitally scanned as Whole Slide Images (WSI). We split the dataset, with 70% as a training set and 30% as a test set. A Deep Learning model was developed for binary classification of mSGBs as positive (FS≥ 1) or negative (FS< 1).
Results: The final dataset comprised of 271 mSGBs of which 153 (56%) had FS< 1 and 118 (44%) had FS≥ 1. In the FS≥ 1 subgroup, 74 (63%) were in the FS=1-2 range, while the remaining biopsies had FS >2 corresponding to the usual FS distribution among SjD population[2]. Our Supervised Learning classifier achieved an AUC=0.932, with a sensitivity of 87% and a specificity of 84%, displaying improvement in performance over previous work[3].
Conclusion: Developing an accurate Artificial Intelligence prediction model, we take a step towards reducing bias in focus score evaluation. Doing so in an automated manner potentially serves to expedite SjD diagnosis, while simultaneously allowing prognostic prediction, via Transfer Learning, which lays the foundations for novel tissue biomarker discovery.
[1]: Sebastian Costa et al. Reliability of histopathological salivary gland biopsy assessment in Sjögren’s syndrome: a multicentre cohort study, Rheumatology, Volume 54, Issue 6, June 2015, Pages 1056–1064, https://doi.org/10.1093/rheumatology/keu453
[2]: Chatzis L, Goules AV, …, Tzioufas AG. A biomarker for lymphoma development in Sjogren’s syndrome: Salivary gland focus score. J Autoimmun. 2021 Jul;121:102648. doi: 10.1016/j.jaut.2021.102648.
[3]: Basseto L et al. OP0232 DEEP LEARNING ACCURATELY PREDICTS FOCUS SCORE AND DIAGNOSIS OF PRIMARY SJÖGREN SYNDROME USING LABIAL SALIVARY GLAND BIOPSIES
Annals of the Rheumatic Diseases 2023;82:152-153.
[4]: Andreas V. Goules, Athanasios G. Tzioufas. “Primary Sjӧgren’s Syndrome: Clinical Phenotypes, Outcome and the Development of Biomarkers.” Autoimmunity Reviews, vol. 15, no. 7, July 2016, pp. 695–703, https://doi.org/10.1016/j.autrev.2016.03.004.
To cite this abstract in AMA style:
Panagiotopoulos K, Tsiknakis N, Zaridis D, Tzioufas A, Fotiadis D, Goules A. Evaluation of Salivary Gland Focus Score in Sjogren’s Disease Using Deep Learning: A Step Towards Tissue Biomarker Discovery [abstract]. Arthritis Rheumatol. 2024; 76 (suppl 9). https://acrabstracts.org/abstract/evaluation-of-salivary-gland-focus-score-in-sjogrens-disease-using-deep-learning-a-step-towards-tissue-biomarker-discovery/. Accessed .« Back to ACR Convergence 2024
ACR Meeting Abstracts - https://acrabstracts.org/abstract/evaluation-of-salivary-gland-focus-score-in-sjogrens-disease-using-deep-learning-a-step-towards-tissue-biomarker-discovery/