深度学习对诊断绩效的影响在良性和恶性甲状腺结节的区分
Esat Kaba1, Merve Solak2, Ayşenur Topçu Varlık2
1Recep Tayyip Erdogan University, Department of Radiology, Rize. esatkaba04@gmail.com.
Medical ultrasonography
|September 4, 2024
概括
深度学习 (DL) 模型从超声波图像中准确地分类甲状腺结节,显著提高了放射科医生的诊断准确性,特别是对于经验较少的从业者.
科学领域:
- 医疗成像医学成像
- 人工智能的人工智能
- 在瘤学瘤学.
背景情况:
- 甲状腺结节很常见,需要准确地区分良性和恶性类型.
- 超声波 (US) 是甲状腺结节评估的主要成像方法.
- 放射科医生的经验会影响甲状腺结节分类的诊断准确性.
研究的目的:
- 评估一种深度学习 (DL) 模型,使用美国图像将甲状腺结节分类为良性或恶性.
- 评估DL辅助对具有不同经验水平的放射科医生的诊断绩效的影响.
主要方法:
- 使用了576张美国甲状腺结节图像的数据集,分为训练 (80%) 和测试 (20%) 集.
- 一个DL模型在结分类培训套上接受了训练.
- 四名放射科医生 (初级住院医生和一名高级放射科医生) 在接受DL模型预测之前和之后对结节进行了分类.
主要成果:
- 在分类甲状腺结节方面,DL模型取得了0.9391的高准确性.
- 随着DL的帮助,放射科医生的精度得到了改善,特别是在初级住院医生 (JR1: 0.7043至0.9130;JR2: 0.7826至0.8696).
- 对所有放射科医生来说,DL辅助提高了精度 (SR:0.8435至0.9304;Srad:0.8522至0.9043).
结论:
- 与个人放射科医生相比,深度学习模型在分类甲状腺结节方面表现出更高的准确性.
- 在所有放射科医生经验水平上,DL辅助显著提高了诊断准确度,在经验较少的住院人员中观察到的最实质性的收益.
相关概念视频
01:27Hyperthyroidism II: Pathophysiology
Hyperthyroidism is a hypermetabolic state caused by elevated levels of thyroid hormones, triiodothyronine (T3) and thyroxine (T4). It results from dysregulation at the thyroid, pituitary, or immune system level and affects multiple organ systems.PathophysiologyThe most common cause of hyperthyroidism is Graves’ disease, an autoimmune disorder in which antibodies, specifically thyroid-stimulating antibodies (TSAb), a subtype of TSH receptor antibodies (TRAb), bind to and activate TSH receptors...
01:27Goiter
Goiter refers to an abnormal enlargement of the thyroid gland that may appear as a diffuse goiter (uniform enlargement) or nodular (single or multiple nodules). Functionally, it is classified as nontoxic (normal/low hormone levels) or toxic (excess hormone production).PathophysiologyDiffuse thyroid enlargement typically results from prolonged stimulation by thyroid-stimulating hormone (TSH) or TSH-like agents, commonly seen in hypothyroidism or iodine deficiency. In contrast, in hyperthyroid...