School of Software, Xinjiang University, Urumqi, 830046, China; People's Hospital of Xinjiang Uyghur Autonomous Region, Xinjiang, China; Xinjiang Key Laboratory of Cardiovascular Homeostasis and Regeneration Research, Xinjiang, China.
School of Computer Science and Technology, Xinjiang University, Urumqi, 830046, China.
Anal Chim Acta. 2024 Nov 22;1330:343302. doi: 10.1016/j.aca.2024.343302. Epub 2024 Oct 4.
Systemic lupus erythematosus (SLE) is a chronic autoimmune disease. Currently, the medical diagnosis of SLE mainly relies on the clinical experience of physicians, and there is no universally accepted objective method for diagnosing SLE. Therefore, there is an urgent need to design an intelligent approach to accurately diagnose SLE to assist physicians in formulating appropriate treatment plans. With the rapid development of intelligent medical diagnostic technology, medical data is becoming increasingly multimodal. Multimodal data fusion can provide richer information than single-modal data, and the fusion of multiple modalities can effectively enhance the richness of data features to improve modeling performance.
In this paper, a cross-modal specific transfer fusion technique based on infrared spectra and metabolomics is proposed to effectively integrate infrared spectra and metabolomics by fully exploiting the intrinsic relationships between features across different modalities, thus achieving the diagnosis of SLE. In this research, a Decision Level Fusion module is also proposed to fuse the representations of two specific transfers further, obtaining the final prediction scores. Comprehensive experimental results demonstrate that the proposed method significantly improves the performance of SLE prediction, with accuracy and Area Under Curve (AUC) reaching 94.98 % and 97.13 %, respectively, outperforming existing methods.
Our framework effectively integrates infrared spectra and metabolomics to achieve a more accurate prediction of SLE. Our research indicates that prediction methods based on different modalities outperform those using single-modality data. The Cross-modal Specific Transfer Fusion module effectively captures the complex relationships within each single modality and models the complex relationships between different modalities.
系统性红斑狼疮（SLE）是一种慢性自身免疫性疾病。目前，SLE 的医学诊断主要依赖于医生的临床经验，尚无普遍接受的客观方法来诊断 SLE。因此，迫切需要设计一种智能方法来准确诊断 SLE，以帮助医生制定适当的治疗计划。随着智能医学诊断技术的快速发展，医学数据正变得越来越多模态化。多模态数据融合可以提供比单模态数据更丰富的信息，并且多模态的融合可以有效地增强数据特征的丰富性，从而提高建模性能。
本文提出了一种基于红外光谱和代谢组学的跨模态特定迁移融合技术，通过充分利用不同模态之间特征的内在关系，有效地融合红外光谱和代谢组学，从而实现 SLE 的诊断。在这项研究中，还提出了一个决策级融合模块，以进一步融合两种特定转移的表示，从而获得最终的预测分数。综合实验结果表明，所提出的方法显著提高了 SLE 预测的性能，准确率和 AUC 分别达到 94.98%和 97.13%，优于现有方法。
我们的框架有效地整合了红外光谱和代谢组学，以实现更准确的 SLE 预测。我们的研究表明，基于不同模态的预测方法优于使用单模态数据的方法。跨模态特定转移融合模块有效地捕捉了每个单一模态内部的复杂关系，并对不同模态之间的复杂关系进行建模。
