24 Nov. 2021
aetherAI, Asia’s leading medical image AI solution provider, today announced CE Mark for aetherAI Hema, for automatic differential counting of bone marrow smears. In collaboration with the National Taiwan University Hospital (NTUH), aetherAI Hema has received the Taiwan Food and Drug Administration (TFDA) approval and Taiwan’s Ministry of Health and Welfare’s approval to commercialize this AI diagnostic tool. This is the first approval case for automatic differential counting on bone marrow smears.
The differential count of blood cells is the basis of diagnostic hematology. In many circumstances, the identification of cells in bone marrow smears is the gold standard for diagnosis. However, morphological assessment and differential count of bone marrow smears are still performed manually. These procedures are tedious, time-consuming, and laden with high inter-operator variation. To solve these issues, aetherAI Hema has been specifically developed to automate the procedure of bone marrow smear differential counting, benefiting all leukemia patients.
According to the WHO International Agency for Research on Cancer, Asia accounts for 54 percent of the world’s leukemia cancer mortality—a percentage much higher than other regions. aetherAI has teamed up with NTUH, the leading medical center in Asia, as well as top experts in hematology disease in Taiwan. NTUH has collected bone smear samples since 1983, and its leukemia patients across Taiwan account for about one-third of these contributions.
Trained on the world’s largest dataset of over 1 million carefully curated cells, aetherAI Hema is able to perform consistently, with a high average accuracy of 94 percent. Each cell will be classified into one of 15 main categories. Results are readily available with just a few clicks.
“aetherAI Hema is the best solution for automatic differential counting of bone marrow smears on the market,” said Dr. Wen-Chien Chou, Department of Laboratory Medicine, National Taiwan University Hospital. “Although some have tried to use AI to solve this issue since 1997, those tools have limitations with certain diseases, low number of annotated cells for datasets, and other variables. aetherAI Hema’s receiving TFDA approval, Taiwan’s highest medical authority, shows its value in clinical practices and meets the needs of hematologists.”
"We’re proud to have obtained this milestone CE Mark for aetherAI Hema Automatic Differential Count of Bone Marrow Smears, the world’s first bone marrow differential AI system. It’s aligned with our mission of developing AI solutions for AI-powered diagnostic support via state-of-the-art technology to elevate the standard of medical imaging diagnosis and improve the quality of care,” said Dr. Joe Yeh, aetherAI co-founder and CEO. aetherAI Hema provides high efficiency and increased consistency of differential count of bone marrow smears. The verified results can be saved in the system and further used for educational purposes or collaborations between medical technicians and hematologists.
Founded in October of 2015, aetherAI is ranked number one in the digital pathology and AI market in Taiwan. Since 2017, aetherAI has been tripling its annual revenue for three years in a row, actively exploring the global market in places like Japan, the Middle East, and the United States. aetherAI-supported research has been published in prestigious international journals. These journals include <Machine Learning Based on Morphological Features Enables Classification of Primary Intestinal T-Cell Lymphomas> by Cancers and <An Annotation-free Whole-slide Training Approach to Pathological Classification of Lung Cancer Types by Deep Neural Network> by Nature Communications.
全球首例獲衛福部、歐盟核准 「骨髓抹片AI分類計數」
臺大醫院與雲象科技宣布，共同研發的「骨髓抹片AI分類計數 aetherAI Hema」已獲衛福部與歐盟CE核准，取得醫材許可證，是該領域全球首例同時獲兩地認證的AI醫材。「骨髓抹片AI分類計數」將一改骨髓抹片細胞人工計數作業，可完成自動分類計數，快速提供量化、可反覆驗證、客觀一致性的數據，協助醫師判讀，提升精準醫療，且可望因取證商轉應用落地，得到大規模導入機會，為血液疾病醫療帶來跨時代的突破。
血液病理進入醫療數位轉型 開展精準判讀新局
根據衛福部統計資料，台灣白血病與骨髓增生性腫瘤病患人數明顯逐年上升，以2016年至2018年患者數量為例，分別是2,168、2,355、2,550；而其中因白血病而死亡的人數每年約1,100人。唯有正確的診斷才能提供最適切的治療，而骨髓抹片之判讀為診斷各種血液疾病的最基本且重要的方法。然而，現行的模式要求，一片骨髓抹片需計數500個血球分類，皆採人工手動，不僅耗時，且細胞計數區域及影像無法存檔紀錄，成果難以驗證。
臺大醫院為台灣血液病診治之重鎮，自1983年來診斷及收藏許多骨髓抹片檢體，為收治白血病病患之重鎮，骨髓抹片判讀的人力素質及數量上在國內外均具優勢。自2018年臺大醫院與專注於醫療影像AI的雲象科技，進行產學合作，率先開發「骨髓抹片AI分類計數」系統。總共用了近60萬個細胞去訓練本AI系統，並以逾兩萬六千個細胞測試，達成可自動分類計數15類骨髓細胞，從原本一張影像依難度不同平均耗時約20分鐘，縮短至5分鐘以內。不僅協助醫師與醫檢師縮短判讀時間，減輕醫療人員負擔，而且提供量化、客觀，可反覆驗證的數據，有助經驗傳承，突破血液疾病臨床診斷與教學研究的瓶頸。
骨髓細胞分類是診斷許多血液疾病的基礎，目前國際市場上尚未有成熟的電腦輔助計數系統。自1997年即有研究者嘗試將AI應用於骨髓細胞的分類計數，截至目前為止，這些研究仍有許多不足之處，例如細胞標註數量不足、僅能辨識骨髓中的少數幾類細胞、只限於應用在某些特定疾病、或是缺乏多中心臨床驗證。這些研究的成果，距離臨床上實用，仍有一段距離。
唯一多國、多中心臨床驗證 符合國際醫療場域需求
臺大醫院與雲象科技合作的「骨髓抹片AI分類計數」屬國際性創新AI應用，國內外均「無類似品」，且骨髓抹片判讀困難，資料量相對稀少，高品質的骨髓抹片較難取得，且分類上較為困難，需要受過高度專業訓練之人員來進行分類，皆為臨床驗證增添挑戰。臺大醫院與雲象科技以高標自許，進行多國、多中心的臨床驗證。254位病人的骨髓抹片分別來自臺大醫院總院、臺大醫院雲林分院、臺北國泰醫院、與美國BioReference Laboratories，一張玻片由兩位醫師及AI標註相互驗證，涵蓋14種骨髓疾病類別，且包括治癒前後的不同臨床病程，跨兩種染色，而此模型的研發資料集，是由臺大醫院血液專科醫師及資深醫檢師進行了超過70萬個細胞標註所組成。
自研發至取證歷經三年努力，2021年10月取得衛福部食藥署及歐盟CE的許可證，驗證此系統在未來運用方面的普遍性，是目前全球最先進的骨髓細胞計數與分類系統，將可推廣於全球的血液實驗室，會是血液疾病診斷的一項革命性的工具與利器。期望從台灣出發，奠基於先進AI技術應用及骨髓細胞型態無人種差異的特色，開拓海外市場。
「骨髓抹片AI分類計數」在臨床驗證、取證過程中得到食藥署諸多諮詢輔導，有助接軌國內外商轉平台，將不同於傳統醫材與資訊硬體的思維，透過產官學的經驗交流，落實在人工智慧、機器學習的智慧醫材推展。雲象科技也願與主管機關協力，推動智慧醫療之新醫材法規典範轉移，使AI應用取證過程更為明確，實現六大核心戰略之健康精準產業的願景。
部分相關媒體報導
Central News Agency https://focustaiwan.tw/sci-tech/202111240022