Development Schedule

Figure 1 shows the development schedule of the integration of GEMINI and NUHS production system.

Figure 2 shows the schematic for AI implementation at NUH. The CDOC and CCDR data from NUHS production system is designed to fed into our GEMINI platform. The different AI modules in GEMINI will provide predicted clinical warnings to the doctors. For the reinforced learning part, doctors will give feedback about the wrongly classified patients and suggest some extra features which can in turn be used for improving the deep learning models.

Figure 3 shows the NUHS DISCOVERY AI Sandbox. Data coming from CDOC and CCDR goes through the NUH FIREWALL and is fed in to the MASTER ACCESS CONTROL layer. On top of the MASTER ACCESS CONTROL layer, many modules are built. These modules include NUH CLINICAL DATABASES module, MASTER VIRTUAL MACHINE module where Forkbase is used as the working storage. PRODUCTIONAL AI MACHINE module is our GEMINI platform. In the PRODUCTIONAL AI MACHINE module, Learning sector corresponds to training phase and Active sector corresponds to inference phase where the results will be aggregated in the OUTPUT AGGREGATOR layer and be shown to the doctors via CPSS and HMS.