利用深度學習進行基於單核苷酸多型性的胃炎型態分類

Abstract

我們的研究旨在探索宿主基因型與胃炎形態之間的關係。為提供客觀的胃炎 診斷和評估，我們採用先進的影像技術和深度學習演算法進行圖像資料的分析。 我們的研究樣本包括 324 位病患，其中 69 位為對照組，255 位被診斷為胃炎。在 考慮年齡和性別的共變量情況下，我們發現了 13 個單核苷酸多型性 (SNPs) 的 p 值低於 10-5 的閾值，其中有 10 個位於第 5 號染色體。這些 SNPs 之間呈現強烈的 連鎖不平衡，顯示它們之間存在著相關性。此外，我們使用胃視鏡影像來訓練深 度學習模型，其中僅以胃竇(antrum)圖像為基礎的模型在 20 個週期後達到了 0.694 的測試準確度，顯示了它在與 SNP 相關的分類任務中具有潛在的應用價值。 這些研究結果指出胃竇影像與我們研究的基因變異可能具有顯著相關性，進一步 凸顯了胃竇特定特徵在理解 SNP 關聯性及其在胃炎中的意義上的重要性。

Our project investigated the relationship between host genotype and the morphology of gastritis, utilizing advanced imaging techniques and deep learning algorithms. We aimed to provide objective diagnosis and assessment of gastritis, thereby enabling precise, personalized medical interventions. Our dataset included 324 patients, of whom 69 were controls, and 255 were diagnosed with gastritis. After adjusting for age and sex, our association analysis revealed 13 significant single nucleotide polymorphisms (SNPs), with 10 located on chromosome 5, suggesting potential genetic correlations with gastritis. Particularly significant were rs10866709 and rs17543640, both located on chromosome 5, with a p-value of 3.2 ×10-6. Furthermore, we trained deep learning models on gastric endoscopy images from various stomach locations, with the model trained on antrum images showing a promising testing accuracy of 0.694 after 20 epochs. This underscored a distinct association between antrum images and SNP-related classification tasks, revealing antrum-specific features' relevance in understanding SNP associations and gastritis implications.