Smalt Christopher J, Sugai Jenna A, Koops Elouise A, Jahn Kelly N, Hancock Kenneth E, Polley Daniel B
Human Health & Performance Systems Group, MIT Lincoln Laboratory, Lexington, MA, USA.
Eaton-Peabody Laboratories, Mass Eye and Ear, Boston, MA, USA.
NPJ Digit Med. 2022 Aug 29;5(1):127. doi: 10.1038/s41746-022-00675-w.
Tinnitus, or ringing in the ears, is a prevalent condition that imposes a substantial health and financial burden on the patient and to society. The diagnosis of tinnitus, like pain, relies on patient self-report, which can complicate the distinction between actual and fraudulent claims. Here, we combined tablet-based self-directed hearing assessments with neural network classifiers to automatically differentiate participants with tinnitus (N = 24) from a malingering cohort, who were instructed to feign an imagined tinnitus percept (N = 28). We identified clear differences between the groups, both in their overt reporting of tinnitus features, but also covert differences in their fingertip movement trajectories on the tablet surface as they performed the reporting assay. Using only 10 min of data, we achieved 81% accuracy classifying patients and malingerers (ROC AUC = 0.88) with leave-one-out cross validation. Quantitative, automated measurements of tinnitus salience could improve clinical outcome assays and more accurately determine tinnitus incidence.
耳鸣，即耳内鸣响，是一种普遍存在的病症，给患者本人及社会带来了沉重的健康和经济负担。与疼痛一样，耳鸣的诊断依赖于患者的自我报告，这可能会使区分真实病例和诈病变得复杂。在此，我们将基于平板电脑的自我听力评估与神经网络分类器相结合，以自动区分耳鸣患者（N = 24）和诈病群体（N = 28），后者被指示假装想象中的耳鸣感知。我们发现两组之间存在明显差异，不仅体现在他们对耳鸣特征的公开报告上，还体现在他们在平板电脑表面进行报告测试时指尖移动轨迹的隐蔽差异上。仅使用10分钟的数据，通过留一法交叉验证，我们对患者和诈病者进行分类的准确率达到了81%（ROC曲线下面积 = 0.88）。对耳鸣显著程度进行定量、自动化测量可以改善临床结果检测，并更准确地确定耳鸣发病率。
