I see You and Computer see You too

The CT-equivalent term, perifissural nodule, has come to refer to a noncalcified solid nodule with sharp margins and a regular shape abutting or near the pleural or fissural margin. Unfortunately, as the authors note, the term perifissural nodule is imprecise and somewhat misleading, in part due to the lack of uniformity of its definition. URL: https://pubs.rsna.org/doi/epdf/10.1148/ryct.2020200409 CT Perifissural lung nodules are typically seen as well-circumscribed, smoothly marginated homogeneous nodules in contact with or closely related to a pulmonary fissure. Most show a triangular, oval/lentiform, rectangular, or dumbbell shape. They may often show a septal attachment to the pleural surface. URL: https://radiopaedia.org/articles/perifissural-lung-nodules

因為工作上接觸過一套肺部 CT 影像後處理的 AI 軟體,  看到這本書就讀一下看作者的想法觀點. 人工智慧種類 - Artificial Narrow Intelligence, Artificial General Intelligence 搜尋, 分類 大資料庫 > 專家系統 - 協助診斷 > AI 軟硬體及環境的進步 使用後的責任歸屬 智慧醫療的價值 台灣有 精緻醫療資料 但是法規沒跟上 學界, 產業 收集資料 電子病歷 提升醫療品質 偵測, 預測 與 預防 生活上的幫手 攔截問題處方簽 可應用的疾病類別 第二意見 - Second opinion  AI 軟體是個工具 可以幫助 醫生, 病人 當下 & 及時 協助 人性化 - 用於照顧病人 智慧病房, 住院品質 慢性病, 長照 個人化醫療 保護隱私權 跟 未來的發展 AI 缺乏情商 #台灣醫療人工智慧創新研究中心 (Artificial Intelligence for Medicine and Health Innovation, AIMHI)  

AAH - Atypical Adenomatous Hyperplasia Atypical adenomatous hyperplasia (AAH) of the lung is a putative precursor lesion of adenocarcinoma of the lung. This entity replaces part of a spectrum of the former bronchoalveolar carcinoma (BAC). Pathology AAHs are represented by localized small cell proliferation, usually measuring ≤0.5 cm, of mildly to moderately atypical type II pneumocytes and/or Clara cells that line along the alveolar walls, alveolar septae, or respiratory bronchiole 3,5,8. Subgrading of those lesions according to the spectrum of cellularity and atypia is not recommended 8. Radiographic features Their radiographic appearances can be somewhat variable but have been commonly described as ground-glass nodules measuring less than 5 mm 3-5, opposed to adenocarcinomas in situ, which are also pre-invasive lesions, that tend to be larger. URL(*1):  https://radiopaedia.org/articles/atypical-adenomatous-hyperplasia-of-the-lung

Depending on the application, unbalance data sample can be a problem.  Here refers to data set used in training AI model. 

The most common type is non-small cell lung cancer (NSCLC).  NSCLC  makes up about 80 to 85 percent of all cases. Thirty percent of these cases start in the cells that form the lining of the body’s cavities and surfaces. This type usually forms in the outer part of the lungs (adenocarcinomas). Another  30 percent  of cases begin in cells that line the passages of the respiratory tract (squamous cell carcinoma). A rare subset of adenocarcinoma begins in the tiny air sacs in the lungs (alveoli). It’s called adenocarcinoma in situ (AIS). This type isn’t aggressive and may not invade surrounding tissue or need immediate treatment. Faster-growing types of NSCLC include large-cell carcinoma and large-cell neuroendocrine tumors. Small-cell lung cancer (SCLC) represents about  15 to 20 percent  of lung cancers. SCLC grows and spreads faster than NSCLC. This also makes it more likely to respond to chemotherapy. However, it’s also less likely to be cured with treatment. In some cases,

AUC or ROC curve - a receiver operating characteristic curve, is a graphical plot that illustrates the diagnostic ability of a binary classifier system as its discrimination threshold is varied.   Binary classification system has four possible outcomes/combination.   Speaking in plain English,    True positive (TP) - equivalent (eqv.) with hit, patient is diagnosed with disease and patient has disease. True negative (TN) - eqv. with correct rejection, patient is diagnosed without disease and patient does not have disease. False positive (FP) - eqv. with false alarm, Type I error.  Patient is diagnosed with disease but in reality does not have one. False negative (FN) - eqv. with miss, Type II error.  Patient is diagnose without disease but in reality does have disease. Or speaking in computer language, Condition positive (P) - the number of real positive cases in the data Condition negative (N) - the number of real negative cases in the data True Positive (TP) - if the