現今呼吸器已經被廣泛的使用於重大手術及重症患者,藉由呼吸器的輔助使病患渡過呼吸衰竭所產生的致命危機,提供較多的機會接受緊急處置以挽救生命。呼吸器適用病患的需求與自主呼吸的能力主要分為侵入型及非侵入型兩種不同的型式,侵入型呼吸器會增加肺炎之發生率7%~41%與死亡率35%~90%,進而延長呼吸器使用天數,導致呼吸器脫離困難。非侵入型是不需氣管插管或氣切的無創傷性機械通氣,可降低肺炎發生率及其它併發症,降低慢性阻塞性肺疾病急性惡化患者插管和死亡率,其他如氣喘、拔管後、手術後、外傷後、肺水腫和等待肺臟移植的病人等,都是可能的受惠者。但在病人心跳血壓不穩定、意識混亂、氣管內痰過多等症狀時,若未能及時使用侵入型呼吸器,反而使病情拖延,甚至病情惡化。所以在侵入型呼吸器及非侵入型呼吸器的選擇評估十分重要。 本研究為輔助評估病人使用非侵入呼吸器的最佳時機,運用人工智慧技術,收集個案醫院病人的生理數據,透過倒傳遞類神經網路、粒子群最佳化演算法及C5.0決策樹,建構最佳預測模型。研究結果顯示,倒傳遞類神經網路結合C5.0決策樹之準確率為93.796%,醫學評估指標面積0.964,皆優於其他模型。倒傳遞類神經網路結合C5.0決策樹建構之規則,經醫師確認在臨床診斷上具有效性且符合文獻,在臨床上對非侵入呼吸器使用時機具有重要的參考價值。
Ventilations have been widely used in major surgeries and on critical illness patients at present time. They help patients through the fatal crisis due to respiratory failure, providing more opportunities for the sick and injured to receive emergency care so more lives can be saved. Ventilations, depending on patients’ needs and their ability to breathe on their own, include two different types—invasive and noninvasive. Invasive ventilations will increase probability of pneumonia from 7% to 41% and mortality from 35% to 90%; they would prolong the days needing ventilations thus leading to difficulty of weaning the dependence. Noninvasive ventilation delivers the mechanically-assisted breaths without the need for intubation, such as endotracheal tube or tracheotomy. It can lower the chance of infections like pneumonia and other complications and can reduce the incubation and mortality rate on patients with acute deterioration of chronic obstructive pulmonary diseases (COPD). For patients with asthma, after extubation, post-surgery, with pulmonary edema, or waiting for lung transplant, noninvasive ventilations can be very helpful. However, for those with symptoms like unstable heart beat and blood pressure, confusion, excess mucus in trachea, etc. it is critical to use invasive ventilations on a timely matter. Without it, symptoms can be delayed and deteriorated. Therefore, the assessment on the selection of invasive or noninvasive ventilations is extremely important. This study aimed to find the best timing to use noninvasive ventilations for patients, using artificial intelligence technology, collecting physiological data of patients in the case hospital, though BPN network, PSO algorithm, and C5.0 decision tree, the most optimal predictive model is constructed. The results show the accuracy of 93.796% in the use of the BPN network combing with C5.0 decision tree, medical evaluation index area at 0.964; both figures are better than the other models. The rule generated from the BPN network combining with C5.0 decision tree, upon confirmation of physicians, has proved most effective in clinical diagnosis and matches with literature results as well. It has important referential values clinically for the timing to use noninvasive ventilations.