Jason 在台北醫學大學給了全院智慧醫療的專題演講


Jason 在台北醫學大學的全院專題演講給了對於智慧醫療的觀察與分享。Jason 從軟體工程師、創業家的角度來切入,資料的收集需要電子電機人才來開發生物感測器,生物訊號(如ECG)裏頭可能會有一些雜訊,需要透過訊號處理來過濾雜訊,來降低誤判率。蒐集到資料後,需要有強力的資料水管工才能完整的把資料往後端送,來做歷史資料大數據的蒐集彙整,等蒐集到足夠的資料且經過專業醫生的標註,來批次的訓練AI模型,ML/AI模型不是光只有程式碼,周邊的配套措施像是特徵提取、資料分析工具、參數的調整紀錄、模型監控的好壞發展、運算資源的設置,都需要一併考慮。智慧醫療需要從系統面來做整體設計規劃,並提到美國iRhythm Technologies所做的心臟貼片ZIO Patch在投給FDA審查時也是以系統來做聲明。在做出系統後,需要與醫生討論應用場景,才有可能找到可行的商業模式,付費的可能是病患、醫院、健康保險單位、或是其他系統廠,新創在設計智慧醫療產品時,需要連未來的商業模式都一併想好。

Jason delivered a speech at Taipei Medical University Hospital. Jason looked at AI healthcare from software engineer and entrepreneur angles. For data collection, it requires electrical engineer to design bio-sensors. In Bio-signal(ex, ECG), there might be some noise. By leveraging signal processing to filter out noise to reduce false alarm. When data is collected, it requires strong data pipeline technology to completely push the data to backend. With enough historical big data and professional medical doctors labeling data, it can do batch processing for building machine learning model. ML/AI model is not just code, there are surrounding supportive things need to be considered. For example, feature extraction, analysis tool, history of configuration parameters tuning, monitoring system, and computation resources setup. For healthcare+AI, it requires total design of the full system. Take iRhythm Techologies as an example, it produces ZIO patch (heart patch). When they filed to get FDA approval, it declares it as a system. When a system was made, it requires doctors' input to know where the system can be put in certain scenario. With certain scenario, a business model can be evaluated. The payer could be patients, hospitals, health insurance company or other system companies. When a startup designs an AI healthcare product, the future business model should also be considered.

s02_Jason_thank_you.jpg s03_Jason_group.jpg s04_Jason_speech2.jpg s05_ai_healthcare.jpg s01_Jason_speech.jpg