Ramathibodi Hospital Collaboration
[2020-01-31] Aiming to create an impact for population health by building trust and relationship with Ramathibodi Hospital
Lessons Learned from working with Ramathibodi Hospital #
- Thai healthcare sector has a lot of room for improvement and they are moving slowly because of internal policies
- In my opinion, one of the greatest obstacles is not using a right man for the job (Doctor has to do IT job)
- The doctors are busy so working with them as a team is quite challenging
- We have to regularly produce output to keep everyone focus on the target outcome
Computer Vision - Chest X-Ray Screening AI #
- We share our Chest X-Ray Screening AI (Detectos®) with Ramathibodi Hospital to integrate the AI with their current workflow and they see benefits and its potential in helping their radiologists prioritize the workload.
- This is the first project and we successfully gain the trust from the radiology department.
Computer Vision - TB/CA Detection #
- We co-author with a resident doctor to research the topic - Diagnostic performance of artificial intelligence system for differentiating between primary lung cancer and mass-like pulmonary tuberculosis
- The model yielded AUPRC 0.891, AUROC 0.940 and 89% accuracy with the test data set.
TB/CA Detection AI Model Training #
- Library: Tensorflow2 Keras
- Architecture: Densenet 121 with pre-trained on a large natural image dataset (Imagenet)
- Last layer activation: Softmax
- Loss function: Categorical Crossentropy with label_smoothing 0.5
- Optimizer: Adam with learning_rate 0.0001
Mobile/Web Application - Ramaai LINE Bot #
- Ramathibodi Hospital has developed a chest x-ray AI model called RAMAAI. It is specialized in COVID-19 detection.
- We develop a LINE Bot application to facilitate the RAMAAI model usage.
- We have learnt more about how to design data collection of the system to comply with Thai PDPA.
- This project is partially funded by PMCU - Program Management Unit for Competitiveness - หน่วยบริหารและจัดการทุนด้านการเพิ่มความสามารถในการแข่งขันของประเทศ (บพข.)
Software Stack #
- PHP8.0 Hyperf+Swoole
- LINE SDK - Messaging API, LINE Login, LINE Front-end Framework (LIFF)