人工智能增强多光谱成像用于烧伤创面评估:英国多中心评估的启示
该多中心前瞻性队列评估了Spectral Deepview设备在烧伤创面评估中的表现,纳入40例患者和70张烧伤图像。AI增强多光谱成像系统与21天临床愈合评估(金标准)相比,敏感度为80.7%,特异度为95.5%,总体准确率为95.3%。
要点
- AI增强多光谱成像预测烧伤愈合的敏感度80.7%,特异度95.5%,总体准确率95.3%。
- 证据仍有限,需进一步评估对临床决策、工作流程和患者结局的影响。
核心结论
- AI增强多光谱成像系统以21天临床愈合评估为金标准,显示出高准确率。
- 该系统结合了诊断精确性、操作效率和便携性。
- 证据基础仍有限,需要进一步研究评估其对临床决策、工作流程整合和患者结局的影响。
研究设计与样本
- 在纽卡斯尔和曼彻斯特成人烧伤中心开展的多中心前瞻性队列评估。
- 纳入标准为18岁及以上、未手术的浅度至全层烧伤患者;入院时进行多光谱成像和临床评估,随访21天。
- 主要结局为愈合预测的可靠性和可重复性,次要结局为设备可行性。
- 使用ImageJ计算AI生成创面图中正确对应金标准分类的像素数,统计分析使用R 4.4.1。
主要发现
- 患者平均年龄51岁,平均总体表面积4.06%,大多数烧伤为烫伤(65.7%)。
- AI增强多光谱成像系统敏感度为80.7%(95% CI: 51.8%-100%)。
- 特异度为95.5%(95% CI: 93.3%-97.8%)。
- 总体准确率为95.3%(95% CI: 93.2%-97.6%)。
临床怎么用
- AI增强多光谱成像可作为烧伤创面愈合评估的辅助工具,提供客观、可重复的输出。
- 设备便携且快速出结果,适合在门诊、手术室和急诊环境中使用。
- 目前证据有限,临床使用时需结合临床判断,并进一步验证其对决策和结局的影响。
局限性
- 证据基础仍有限,需要进一步开发。
- 需未来研究评估设备对临床决策、工作流程整合和患者结局的影响。
- 原文未报告其他局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
INTRODUCTION: Accurate burn wound assessment is essential for effective treatment, yet it remains heavily dependent on clinical judgment, which is highly subjective. While various optical-based instruments have been developed to address this issue, their clinical effectiveness has been limited due to high cost, penetration depth, lack of portability and validation primarily for use on days 2-5 post injury. The integration of artificial intelligence (AI) with multispectral imaging (MSI) represents a potential advancement in enhancing the accuracy and consistency of wound assessment. MSI captures complex physiological data beyond the visible spectrum, while AI may facilitate its interpretation by generating objective, consistent, and reproducible outputs. However, this relatively new approach still lacks robust clinical validation due to its novelty and the limited body of supporting evidence. This product evaluation examines the first application of AI-enhanced MSI for burn wound assessment in a multi-centre UK setting. METHOD: We conducted a multicentre prospective cohort evaluation of the Spectral Deepview device at the Newcastle upon Tyne and Manchester Adult Burn Centres. Inclusion criteria were patients 18 years old and older with superficial to full thickness burns that had not undergone surgery. MSI and clinical assessment were performed on admission, and the patients followed up for 21 days. The primary outcome was the reliability and reproducibility of healing prediction, whilst the secondary outcome was the instrument's feasibility. The AI's prediction was compared to the clinical healing assessment at 21 days, which served as the ground truth. ImageJ was used to calculate the number of pixels in the AI-generated wound maps that correctly corresponded to the ground truth classification of healed and unhealed areas, enabling a quantitative comparison. Statistical analysis was performed using the statistical package R (version 4.4.1). RESULTS: The clinical evaluation included 40 patients and 70 burn images, generating approximately 13 million data points. The mean age of the patients was 51, with an average Total Body Surface Area (TBSA) of 4.06 %. Most burns were scalds (65.7 %). The AI-enhanced MSI system demonstrated a sensitivity of 80.7 % (95 % CI: 51.8 %-100 %) and a specificity of 95.5 % (95 % CI: 93.3 %-97.8 %). The overall accuracy of the system was 95.3 % (95 % CI: 93.2 %-97.6 %). The mean time from scan to result was five minutes and twelve seconds. The device was portable and effectively utilised in clinics, operating theatres, and emergency departments. CONCLUSION: Our clinical evaluation demonstrates that the AI-enhanced MSI system offers high accuracy compared to clinical healing outcomes as the ground truth. It combines diagnostic precision, operational efficiency, and portability. However, the evidence base remains limited and requires further development. Future research is needed to assess the device's impact on clinical decision-making, workflow integration, and patient outcomes.
引用
- Tan P, Nyeko-Lacek M, Walsh K, et al. Artificial intelligence-enhanced multispectral imaging for burn wound assessment: Insights from a multi-centre UK evaluation. Burns. 2025. PMID: 40858093. doi:10.1016/j.burns.2025.107550
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。