Development of an AI-driven algorithm based on acute diagnostic CT brain scans to predict recovery from intracerebral haemorrhage

脑出血(ICH)是全球范围内发病率和死亡率的主要原因。与缺血性中风不同，脑出血是一种占位性病变，可在发病后的最初几个小时到几天内导致肿块效应、突出综合征和死亡。手术可以通过减少质量效应来防止死亡，但并没有令人信服地改善幸存者的康复。出血的位置与其他重要结构的关系可能对手术后恢复的可能性至关重要。皮质脊髓束的完整性对于缺血性卒中患者的运动恢复很重要，我们假设保存的皮质脊髓束也将预示着脑出血患者良好的运动恢复，从而预示着手术的反应，否则患者可能会死于质量效应。我们的首要目标是开发一种基于常规诊断性CT脑扫描的实用、可靠和有效的工具，以改善急性脑出血的预后预测，并测试这是否有助于识别可能从手术中受益的患者。现有的确定皮质脊髓束完整性的金标准方法是磁共振扩散张量成像和经颅磁刺激，但对于病情严重的脑出血患者来说，这两种方法都非常具有挑战性。然而，所有患者都接受了急性CT脑成像，CT提供了关于出血和周围积水所涉及的结构的丰富且大部分未被利用的信息。这项研究将应用计算机视觉和机器学习到现有的大型脑出血数据集中的CT扫描，以确定CT是否可以可靠地预测脑出血后皮质脊髓束的完整性和/或良好的长期恢复。我们将使用现有的约500名急性脑出血患者的CT脑扫描和MR扩散张量成像的数据集，初步确定诊断性CT脑扫描是否可以预测皮质脊髓束的完整性。利用现有的临床试验数据档案/数据集，总共约6,000名患者可进行诊断CT脑扫描，我们将进一步开发我们的算法，基于修正的Rankin评分预测90天后恢复良好。最后，我们将测试该算法是否可以使用现有的诊断性CT脑扫描来识别最有可能从手术中受益的患者，这些患者来自以前的脑出血手术试验中的大约2000名患者。然后，我们将与Brainomix合作，开发这一工具，作为eStroke Suite的一部分，用于常规临床使用。将计划进行进一步的神经外科临床试验，部署我们的ICH CT分析工具来选择招募的患者。

Intracerebral haemorrhage (ICH) represents a major cause of morbidity and mortality on a global scale. Unlike ischaemic stroke, ICH is a space-occupying lesion and can lead to mass effect, herniation syndromes and death in the first hours to days after onset. Surgery can prevent death by reducing mass effect but has not convincingly improved recovery in survivors. The location of the haemorrhage in relation to other vital structures is likely to be critical to the potential for recovery after surgery. Integrity of the corticospinal tract is important for motor recovery in ischaemic stroke and we hypothesise that a preserved corticospinal tract will also predict a good motor recovery in ICH and hence response to surgery, in a patient who might otherwise die of mass effect. Our overarching aim is to develop a practical, reliable and validated tool based on routine diagnostic CT brain scans to improve outcome prediction in acute ICH and test whether this may help identify patients likely to benefit from surgery. Existing gold-standard methods to determine corticospinal tract integrity are MR diffusion tensor imaging and transcranial magnetic stimulation, but both can be very challenging in acutely unwell ICH patients. However, all patients undergo acute CT brain imaging and CT provides rich and largely untapped information about the structures involved by the haemorrhage and surrounding oedema.This studentship will apply computer vision and machine learning to CT scans from existing, large ICH datasets to determine whether CTs can reliably predict corticospinal tract integrity and/or a good long-term recovery after ICH. We will initially establish whether diagnostic CT brain scans can predict integrity of the corticospinal tract using existing datasets of around 500 acute ICH patients with both CT brain scans and MR diffusion tensor imaging. Using existing clinical trial data archives/datasets with a total of around 6000 patients with diagnostic CT brain scans available, we will further develop our algorithm to predict good recovery at 90 days based on the modified Rankin Scale score. Finally, we will test whether the algorithm can identify patients most likely to benefit from surgery using existing diagnostic CT brain scans from around 2000 patients from previous ICH surgery trials.Working with Brainomix, we will then develop this tool for routine clinical use as part of the eStroke Suite. Further clinical trials of neurosurgery will be planned, with our ICH CT analysis tool deployed to select patients for recruitment.