From the Division of Plastic and Reconstructive Surgery, Massachusetts General Hospital, Harvard Medical School.
Plast Reconstr Surg. 2023 Feb 1;151(2):405-411. doi: 10.1097/PRS.0000000000009855. Epub 2022 Nov 15.
Recent evidence has shown that patient drawings of pain can predict poor outcomes in headache surgery. Given that interpretation of pain drawings requires some clinical experience, the authors developed a machine learning framework capable of automatically interpreting pain drawings to predict surgical outcomes. This platform will allow surgeons with less clinical experience, neurologists, primary care practitioners, and even patients to better understand candidacy for headache surgery.
A random forest machine learning algorithm was trained on 131 pain drawings provided prospectively by headache surgery patients before undergoing trigger-site deactivation surgery. Twenty-four features were used to describe the anatomical distribution of pain on each drawing for interpretation by the machine learning algorithm. Surgical outcome was measured by calculating percentage improvement in Migraine Headache Index at least 3 months after surgery. Artificial intelligence predictions were compared with clinician predictions of surgical outcome to determine artificial intelligence performance.
Evaluation of the data test set demonstrated that the algorithm was consistently more accurate (94%) than trained clinical evaluators. Artificial intelligence weighted diffuse pain, facial pain, and pain at the vertex as strong predictors of poor surgical outcome.
This study indicates that structured algorithmic analysis is able to correlate pain patterns drawn by patients to Migraine Headache Index percentage improvement with good accuracy (94%). Further studies on larger data sets and inclusion of other significant clinical screening variables are required to improve outcome predictions in headache surgery and apply this tool to clinical practice.
最近的证据表明，患者的疼痛绘图可以预测头痛手术的不良结果。鉴于疼痛绘图的解释需要一些临床经验，作者开发了一个能够自动解释疼痛绘图以预测手术结果的机器学习框架。该平台将使临床经验较少的外科医生、神经科医生、初级保健医生，甚至患者能够更好地了解头痛手术的适应证。
一项随机森林机器学习算法在 131 张前瞻性头痛手术患者提供的疼痛绘图上进行了训练，这些患者在接受触发点去激活手术前进行了训练。24 个特征用于描述每个绘图上疼痛的解剖分布，以便机器学习算法进行解释。手术结果通过计算手术后至少 3 个月偏头痛头痛指数的百分比改善来衡量。人工智能预测与临床医生对手术结果的预测进行比较，以确定人工智能的性能。
对数据测试集的评估表明，该算法的准确性（94%）始终高于经过训练的临床评估者。人工智能将弥漫性疼痛、面部疼痛和顶点疼痛加权为预测手术结果不良的强预测因素。
这项研究表明，结构化算法分析能够以较高的准确性（94%）将患者绘制的疼痛模式与偏头痛头痛指数百分比改善相关联。需要进一步研究更大的数据集和纳入其他重要的临床筛选变量，以提高头痛手术的预后预测，并将该工具应用于临床实践。
