基于百度ai的抑郁症分析
By Tracie White
由崔西怀特
Current methods used to diagnose and treat depression are imprecise at best, relying largely on subjective answers to survey questions, said Leanne Williams, PhD, Stanford professor of psychiatry. At worse, these approaches can result in treatment choices that further postpone a patient’s recovery as the disease progresses.
斯坦福大学精神病学教授Leanne Williams博士说，目前用于诊断和治疗抑郁症的方法充其量是不精确的，很大程度上依赖于对调查问题的主观回答。 更糟糕的是，这些方法会导致选择治疗，随着疾病的进展，进一步推迟患者的康复。
“Currently treatments are a trial and error process,” Williams said. “It’s one size fits all. If the first treatment doesn’t work, a second gets tried. We need a more precise tool to pick the best treatment option first.”
“目前的治疗是一个反复试验的过程，”威廉姆斯说。 “这是一种适合所有人的尺码。 如果第一种治疗无效，则尝试第二种治疗。 我们需要一种更精确的工具来首先选择最佳治疗方案。”
Williams and her collaborators set out to define a more effective model, which they hope can be used in clinics soon. In a recent study, they deployed an algorithm to interpret brainwave patterns unique to individuals with depression, with the goal of better pinpointing which symptoms change with treatment.
威廉姆斯和她的合作者着手定义一个更有效的模型，他们希望该模型能很快用于临床。 在最近的一项研究中 ，他们部署了一种算法来解释抑郁症患者独有的脑电波模式，目的是更好地查明哪些症状会随治疗而改变。
“We know that depression is very heterogenous, and that there are at least 1,000 unique combinations of symptoms that can be diagnosed as depression,” said Williams, who is the director of Stanford’s Center for Precision Mental Health and Wellness. “We’ve found that brainwave measurements can be used to help identify which particular symptoms change with antidepressant treatment and which do not.”
斯坦福大学精确心理健康与健康中心主任威廉姆斯说：“我们知道抑郁症是非常不同的，并且至少有1000种独特的症状组合可以被诊断为抑郁症。” “我们发现，脑电波测量可用于帮助确定抗抑郁药治疗会改变哪些特定症状，而哪些不会。”
预测哪些症状会改善 (Predicting which symptoms improve)
Major depression is the most common mental disorder in the United States, affecting about 7% of adults in 2017, according to the National Institute of Mental Health. Among those, about half never get diagnosed; and for those who do, finding the right treatment can take years with the current trial and error process.
根据国家心理健康研究所的数据，严重抑郁症是美国最常见的精神障碍，2017年影响了约7％的成年人。 其中，约有一半从未被诊断出。 对于那些这样做的人，在当前的反复试验过程中，找到正确的治疗方法可能需要花费数年的时间。
For Williams’ study, data was collected from 518 patients diagnosed with depression randomized to eight weeks of treatment with one of three different antidepressants. Based on brainwave data, the algorithm successfully predicted which symptoms improved with treatment, with highest performance for seven symptoms including insight and loss of weight.
对于Williams的研究，收集了518名被诊断为抑郁症的患者的数据，这些患者被随机分配到使用三种不同抗抑郁药之一治疗八周。 基于脑电波数据，该算法成功预测了哪些症状会随着治疗而改善，其中七个症状(包括洞察力和体重减轻)的表现最高。
Creating new, objective, high-tech lab tests to help diagnose mental disorders has long been the goal of Williams and other translational neuroscientists. Instead of running blood tests or using measurements taken from heart monitors, clinicians currently rely on a survey: Either the patient or the physician lists the symptoms themselves. If a patient has a certain number of a wide variety of different symptoms — among them low mood, appetite changes, loss of insight, loss of energy and poor concentration — they receive a broad diagnosis of clinical depression.
长期以来，创建新的，客观的高科技实验室测试以帮助诊断精神疾病一直是Williams和其他翻译神经科学家的目标。 目前，临床医生没有进行血液测试或使用心脏监护仪进行的测量，而是依靠一项调查：患者或医生自己列出症状。 如果患者有一定数量的各种各样的不同症状-其中包括情绪低落，食欲不振，失去知觉，精力不足和注意力不集中-他们将得到广泛的临床抑郁症诊断。
数据中的复杂关系 (Complex relationships in data)
For the new model, Williams collaborated with researchers at Stanford’s AI for Healthcare Bootcamp in a group led by Andrew Ng, PhD, adjunct professor of computer science. The team set out to design an algorithm able to predict improvement of various depressive symptoms with antidepressant treatment. Individual symptom data were combined with individual recordings from electroencephalography (EEG) tests which monitored electrical activity in the brains of the participants.
对于新模型，威廉姆斯与斯坦福大学AI医疗保健训练营的研究人员合作，由计算机科学副教授Andrew Ng博士领导。 该团队着手设计一种算法，该算法能够预测抗抑郁药治疗后各种抑郁症状的改善。 个体症状数据与脑电图(EEG)测试的个体记录相结合，脑电图(EEG)测试监测参与者大脑中的电活动。
“We can apply artificial intelligence to learn complex relationships in data,” said Pranav Rajpurkar, a PhD student in computer science, who shared lead authorship of the study with Jingbo Yang, a master’s student. “We are able to learn and discover interesting relationships between a patient’s depression symptoms — and EEG readings — at start of treatment, and their depression symptoms eight weeks in.”
“我们可以将人工智能学会数据中的复杂关系，”说Pranav Rajpurkar ，博士生在计算机科学，谁分享了该研究的主要作者颈脖杨 ，硕士生。 “我们能够学习并发现患者在开始治疗时的抑郁症状以及脑电图读数与八周内的抑郁症状之间的有趣关系。”
The algorithm was also able to identify individuals with clinical symptoms associated with a higher risk of poor outcomes, such as suicide, Rajpurkar said. These symptoms may otherwise have been missed due to the subjective nature of diagnosis. For example, the symptom labeled ‘poor insight’, which means that the patient may not be able to realize the extent of their illness, often gets overlooked.
Rajpurkar说，该算法还能够识别出临床症状与不良后果风险较高(例如自杀)相关的个体。 由于诊断的主观性质，可能会忽略这些症状。 例如，标有“洞察力差”的症状通常意味着人们可能无法意识到自己的病情。
“We need new models, such as this one, to provide objective measures of these depression risk factors to identify people who may benefit from more intensive treatments, or treatments other than antidepressants, with the goal of getting the best treatment fast,” she said.
她说：“我们需要诸如此类的新模型，以提供这些抑郁症危险因素的客观指标，以识别可能受益于更深入的治疗或除抗抑郁药以外的治疗方法的人，以期尽快获得最佳治疗。” 。
Image by Andrii
图片由Andrii提供
Originally published at https://scopeblog.stanford.edu on June 24, 2020.
最初于 2020年6月24日 发布在 https://scopeblog.stanford.edu 。
基于百度ai的抑郁症分析
