视网膜色素变性深度学习模型研究进展及应用展望

Research Progress and Application Prospects of Deep Learning Models for Retinitis Pigmentosa

摘要: 视网膜色素变性(Retinitis Pigmentosa, RP)是常见的遗传性视网膜疾病，以进行性感光细胞凋亡为特征，最终导致不可逆的视力丧失。近年来，深度学习模型的快速发展为RP的自动诊断和进展预测提供了突破性的技术路径。本文系统梳理了RP深度学习领域的研究进展，从数据资源建设、关键技术演进、多模态影像分析、视功能预测等维度展开评述。研究表明，基于CNN和Vision Transformer的RP模型在疾病分类(AUC达0.94)和视功能预测(准确率超过80%)等方面已取得较好的预测性能。然而，由于RP疾病罕见、遗传异质性高、标注数据集匮乏，模型泛化能力与临床可解释性仍面临重大挑战。基于此，本文进一步展望了自监督学习、生成式模型与大模型预训练等前沿技术在推动RP精准诊疗中的潜力。

Abstract: Retinitis Pigmentosa (RP) is a common inherited retinal disorder characterized by progressive apoptosis of photoreceptor cells, ultimately leading to irreversible vision loss. In recent years, the rapid development of deep learning models has provided a breakthrough technical pathway for automated diagnosis and progression prediction of RP. This article systematically reviews the research progress of deep learning in the field of RP, covering aspects such as data resource construction, key technological evolution, multimodal image analysis, and visual function prediction. Studies have shown that RP models based on CNN and Vision Transformer have achieved good predictive performance in disease classification (AUC up to 0.94) and visual function prediction (accuracy exceeding 80%). However, due to the rarity of RP, high genetic heterogeneity, and scarcity of annotated datasets, model generalizability and clinical interpretability remain major challenges. Based on this, this article further discusses the potential of cutting-edge technologies such as self-supervised learning, generative models, and large model pre-training in promoting precision diagnosis and treatment of RP.