周读论文系列笔记(1)-review-Artificial intelligence in healthcare

第一次…不知道怎么写，好纠结(꒦_꒦)

原文链接：https://www.nature.com/articles/s41551-018-0305-z

文章目录

• A comprehensive perspective on AI in medicine

• The latest advancements in AI technologies and their biomedical applications

• • Recent advancements in image-based diagnostic methods*
    • • Advanced radiological imaging techniques*
      • Radiological imaging techniques are essential for analyzing internal body structures.
      • Skin condition analysis is a critical component of dermatology.
      • Eye science research focuses on diagnosing ocular conditions.
      • Pathological analysis is fundamental to advancing pathology studies.

• 2.2 基因组解析

  • 机器学习在生物标志物识别中的应用 研究生物标志物
  • 临床结果预测与患者随访 进行临床结果预测与患者的监控
  • 可穿戴设备辅助下的健康状态推断 借助可穿戴设备进行健康状况推断
  • 自动化医疗手术 自动式或自动化的医疗手术系统
• 3. Obstacles encountered during AI development
• 4. Various social, economic, and legal obstacles

1.A historical overview of AI in medicine

(1) 初代 AI 系统：临床决策支持系统（20 世纪中期）
基于规则的方法用于分析心电图、诊断疾病、制定治疗方案等。
(2) 机器学习
(3) 深度学习

Recent advancements in AI technologies have made remarkable contributions to biomedical applications.

2.1 Image-based diagnosis 基于图像的诊断

Nowadays, automated medical image diagnosis is generally regarded as the most remarkable field within medical AI applications.

2.1.1 Radiology 放射学

Applying medical imaging techniques enables the detection and diagnosis of diseases. Medical imaging techniques include X-ray radiography, computed tomography, MRI, and positron emission tomography. Radiological practice predominantly utilizes imaging methods for the purpose of diagnosis.

Thanks to advancements in modern machine learning techniques, numerous AI-driven applications in the field of radiology have achieved expert-level accuracy. These include computerized tomography (CT) scans for identifying lung nodules, X-ray diagnostics for pulmonary tuberculosis, analysis of chest X-rays for common lung diseases, and mammogram-based breast mass recognition.

These studies utilized a technique referred to as transfer learning, by borrowing previously established deep neural networks trained on millions of natural, non-medical images and fine-tuning the neural network connections using thousands of biomedical images.

Governmental approval: The FDA approved a deep learning-based system for the diagnosis of cardiovascular diseases using cardiac MRI images in 2018.

2.1.2 Dermatology 皮肤病学

典型的皮肤黑色素瘤。
色素性肿瘤 ABCED 法则。
用于分类良性与恶性病变的相片，
CNN训练于129,450临床图像上以达到皮肤癌诊断水平与皮肤学家相当的准确性。

2.1.3 Ophthalmology 眼科学

Fundus photography（眼底照相）is a valuable diagnostic tool capable of detecting early signs of various diseases such as diabetic retinopathy（糖尿病视网膜病变）, glaucoma（青光眼）, retinal tumors（视网膜肿瘤）, and age-related macular degeneration（年龄相关性黄斑疾病）。It plays a crucial role in identifying risk factors for preventable blindness.

2.1.4 Pathology 病理学

The histopathological examination serves as the benchmark for diagnosing numerous types of cancers.
AI-powered convolutional neural networks (CNNs) enable automated detection of prostate cancer from biopsy specimens. Early-stage detection of breast cancer metastasis within lymph nodes is achievable via these networks. Additionally, they facilitate the diagnostic evaluation of mitotic patterns in breast tumor samples.

2.2 Genome interpretation 基因组解释

Advanced sequencing techniques produce vast quantities of raw genomic data, which are essential for comprehensive genomic research.
Deep neural networks are capable of annotating pathogenic variants of genetic DNA while determining their functional roles in non-coding regions.

2.3 Machine learning for biomarker discovery 生物标志物的发现

The identification of biomarkers relies upon pinpointing previously undetected associations between an extensive array of measurements and various phenotypes.
Machine learning possesses the capability to identify molecular patterns linked to disease states and subtypes, take into account high-order interactions among measurements, and generate omics-based signatures to predict disease outcomes.
Cancers, infectious diseases, and the risk of Down’s syndrome.

通过临床效果预判实现患者的全程管理

基于电子健康记录（EHR）预测临床结果的方法能够提供精准的数据支持

2.5 Determining health indicators via wearable technology 通过可穿戴技术确定健康指标

2.6 Autonomous robotic surgery 自主机器人手术

3.Technical challenges in AI developments

4.Social, economic and legal challenges