成果亮點

Feasibility of pretreatment FDG PET radiomics in predicting circumferential resection margin involvement for esophageal cancer after neoadjuvant concurrent chemoradiotherapy

Background

Involvement of circumferential resection margin (CRM) proves to be a valuable factor in predicting prognosis for patients with esophageal cancer. This study aimed to explore the feasibility of using pretreatment 18F-fluorodeoxyglucose positron emission tomography/computed tomography (FDG PET/CT) in predicting CRM involvement for patient received neoadjuvant concurrent chemoradiotherapy (CCRT).

Methods

We retrospectively enrolled patients with esophageal cancer who received neoadjuvant CCRT followed by radical surgery between October 2013 and July 2023. All patients received FDG PET/CT examinations before CCRT. The maximum standardized uptake value (SUVmax) within a tumor was identified, and the metabolic tumor volume (MTV) was calculated by using a fixed SUV threshold method (50 % of SUVmax). Descriptive features and five groups of radiomic features (Gray Level Co-occurrence Matrix [GLCM], Gray Level Dependence Matrix [GLDM], Gray Level Run Length Matrix [GLRLM], Gray Level Size Zone Matrix [GLSZM], and Neighboring Gray Tone Difference Matrix [NGTDM]) were further calculated to describe the heterogeneity of the FDG uptakes within MTV. The CRM involvement status was defined by the Royal College of Pathologists (RCP). Tumor presence at or within 1 mm of the cut margin was considered CRM positive. Finally, we used the area under the receiver operating characteristic curve (AUC) to evaluate the predictive performance of radiomic features for CRM. The significance level was set to 0.05.

Results

A total of 71 cases of esophageal tumors, encompassing data from 70 patients were included. Of the 70 patients, 94% were male with a mean age 58 years across the entire cohort. Sixteen tumors were proved as CRM positive (16/71, 22.5%). Finally, 2 descriptive features and 10 radiomic features were selected that can effectively predict CRM status. These ten radiomic features exhibit robust predictive power for CRM status, demonstrating significance across various discretization bin number. The best performing feature is Skewness (AUC=0.716, p-value<0.0001).

Conclusions

This preliminary result revealed that CRM involvement for esophageal cancer tumors could be predicted by using radiomic features. Future studies are warranted for clinical application.

Predictors of the survival for platinum-refractory head and neck squamous cell carcinoma by using contrast-enhanced Magnetic Resonance Imaging

使用對比增強磁振造影（MRI）預測對鉑類化合物無效之頭頸鱗狀細胞癌患者的存活率

簡介

對鉑類化合物無效（Platinum-refractory）的頭頸鱗狀細胞癌患者其預後相當不佳。即便使用免疫檢查點療法（Immune Checkpoint Therapy），其整體存活期（Overall Survival, OS）仍然只有 7.7 個月。

方法

研究對象為對鉑類化合物無效之頭頸鱗狀細胞癌患者，收集患者於接受治療前的 T1 權重之對比增強磁振造影（T1-Weighted Contrast-Enhanced Magnetic Resonance Imaging, T1 CE MRI），邀請兩名專科醫師獨立進行影像判讀，標記腫瘤復發的位置。接著計算 axial、coronal 以及 sagittal 三個平面腫瘤位置的交接點，以此座標為圓心劃定半徑為 25 個像素（pixel）的區域（Region of Interest, ROI），並排除空氣。最後透過分析 ROI 的描述性特徵與四組紋理特徵，呈現 MRI 強度的分佈與異質性。

結果與結論

共回溯性納入 30 位頭頸癌患者，其中 19 位的一線治療方式為免疫檢查點治療，11 位則接受西妥昔單抗化療（Cetuximab-Based Chemotherapy）。在治療結束後的追蹤期間共有 17 位患者死亡，OS 之中位數為 7.0 個月。研究結果發現 ROI 的面積大小與死亡並沒有顯著相關，而有四項紋理特徵具有預測死亡的能力，分別是 High Gray-level Run Emphasis (HGRE)（AUC=.747, p=.023）、Short-Run High Gray-level Emphasis (SRHGE) of GLRLM（AUC=.842, p=.002）、High Gray-level Zone Emphasis (HGZE)（AUC=.817, p=.003）以及 Short-Zone High Gray-level Emphasis (SZHGE) of GLSZM（.833, p=.002）。

Artificial intelligence-based healthcare-associated bloodstream infection surveillance and classification system

基於人工智能的健康照護相關血流感染監測及分類系統

簡介

CLABSI 是一種原發性的血流感染，是放置中心導管過程中因微生物移至導管外表面或輸注液體而導致血流感染。中心導管相關血流感染（Central line-associated bloodstream infections, CLABSI）可能導致病患住院時間延長、醫療費用增加，或死亡風險增加。然而，辨別患者是否發生中心導管相關血流感染需耗費大量的時間與人力，辨別的效率取決於感染控制人員的專業知識。本研究期望能以人工智能（Artificial intelligence, AI）技術結合網頁系統，自動監測中心導管相關血流感染，藉以改善醫院流程，提升醫療品質。

方法

我們開發了一個網頁系統，彙整與重構來自醫院資訊系統中各個子系統的臨床資料，將過往以事件為導向轉換為以病人為中心的資料架構。首先，將導管記錄中非標準化格式的資料進行標準化，特別針對中心導管的名稱、放置位置，以及用途。接著，應用自然語言處理來辨識檢驗報告中的菌種以及護理記錄中的生命徵象與症狀。最後，開發一項以規則主導（rule-based）的 AI 模組，基於結構化數據（例如生命參數）等，以及台灣疾病管制署於 2018 年所發布的監測定義，進行中心導管相關血流感染的判定。透過與感控人員判定結果的比較，評估 AI 模組的判定能力，對於兩者不一致的部分將進行同儕審查，以驗證其準確性並確保偵測能力。

結果與結論

研究共納入 9,173 名住院患者，22,961 份陽性血液樣本；其中有 299 份樣本（1.3%）被感控人員判定為 CLABSI。而在 299 份樣本中，AI 判定為 CLABSI 的為 291 份（true positive, TP），8 份為非 CLABSI（false negative）。相反地，有 272 份不被感控人員判定為 CLABSI 的樣本被 AI 歸類到 CLABSI（false positive），在與感控人員確認之後，將其中 73 份原先被人工判定為非 CLABSI 的樣本修正為 CLABSI。最終 AI 模組的敏感度、特異性、陽性預測值和陰性預測值分別為 97.85%、99.11%、64.65 % 和 99.96%。