人工智能读片技术在肺部转移瘤患者综合诊断和治疗中的应用

Application of artificial intelligence in comprehensive diagnosis and therapy of patients with pulmonary metastases

目的应用人工智能(AI)读片技术明确肺部转移瘤的数量和位置,制订合适的手术方案,改善患者预后。方法收集2014年1月—2020年1月在上海交通大学医学院附属仁济医院胸外科行手术治疗的169例肺部转移瘤患者术前和术后随访的影像学资料并进行对比,统计经过AI读片后考虑为肺部转移瘤的结节、手术中切除的结节、未被切除但AI读片后考虑为肺部转移瘤的结节,判定术后“复发”的结节是否为原来影像学资料上遗漏或存疑的结节。采用单因素和多因素分析影响患者预后的危险因素。结果169例患者中,有159例患者每年进行随访直至随访,终点事件发生。肿瘤发生肺部进展27例,因术前读片遗漏导致肿瘤进展13例(48.15%,13/27),其中仅为术前读片遗漏6例、术前读片遗漏+新发肺部转移瘤7例;新发肺部转移瘤20例(74.07%,20/27),其中仅为新发肺部转移瘤13例、术前读片遗漏+新发肺部转移瘤7例;转移瘤切缘复发2例(7.40%,2/27)。术前读片未遗漏患者术后1、2、3年无进展生存率(PFS)均显著高于术前读片遗漏者(P值均<0.05)。单因素分析结果显示,肿瘤是否系寡转移[风险比(HR=4.636,95%CI为4.201~5.117]、原发肿瘤病理类型(HR=1.373,95%CI为1.312~1.438)、肺部转移瘤是否被完整切除(HR=1.632,95%CI为1.580~1.686)为影响患者预后的危险因素(P值均<0.05)。多因素分析结果显示,肿瘤是否系寡转移(HR=3.211,95%CI为2.839~3.565)、原发肿瘤病理类型(HR=1.339,95%CI为1.279~1.402)、肺部转移瘤是否被完整切除(HR=1.485,95%CI为1.439~1.536)是影响患者预后的危险因素(P值均<0.05)。结论采用AI读片技术协助人工读片有助于检测出更多的肺部转移瘤,可指导临床医师制订综合诊断和治疗方案,改善患者预后。

Objective To formulate a sargical strategy by using artificial intelligence(AI)technology to clarify the number and location of pulmonary metastases,so as to improve patient outcomes.Methods Imaging data of 169 patients with pulmonary metastases who underwent thoracic surgery in our hospital from January 2014 to January 2020 were collected.Preoperative and postoperative follow-up imaging data were compared.The nodules considered as pulmonary metastases after AI scanning,resected nodules,and the nodules that were not removed but considered to be pulmonary metastases after AI scanning were recorded.Then the“relapsed”nodules were identified whether had been missed or suspected nodules in the original imaging data.Univariate and multivariate analyses were used to analyze the relationship between the complete resection of lung metastases and the prognosis of patients,as well as the risk factors for the prognosis of patients with lung metastases.Results Of them,159 patients were followed up every year until the endpoint event.The tumor progressed in 27 cases.There were 13 cases of tumor progression due to omission of preoperative reading(48.15%,13/27,only omission of preoperative reading in 6 cases,and omission of preoperative reading and new lung metastasis in 7 cases).There were 20 cases of new lung metastases(74.07%,20/27),including 13 cases of new metastatic tumors and 7 cases of both omission of preoperative reading and new metastatic tumors.The recurrence of resected margin of metastatic tumor was found in two case(7.40%,2/27).The one-,two-,and three-year progression free survival(PFS)of patients with incompletely resected pulmonary metastases were significantly shorter than those with complete resection(all P<0.05).Univariate analysis showed that oligometastasis(HR=4.636,95%CI:4.201-5.117),primary tumor pathology type(HR=1.373,95%CI:1.312-1.438),and incomplete resection(HR=1.632,95%CI:1.580-1.686)were risk factors for the prognosis(P<0.05).Multivariate analysis also showed oligometastasis(HR=3.211,95%CI:2.839-3.565),primary tumor pathology type(HR=1.339,95%CI:1.279-1.402),and incomplete resection(HR=1.485,95%CI:1.439-1.536)were risk factors for the prognosis(all P<0.05).Conclusion Artificial intelligence technology can identify more pulmonary metastases,contributing to comprehesive diagnosis and treatment,and improving the prognosis of patients.