Evidence based learning, and decomposition of questions to formulate stronger hypothesis.
Discussed the Jeopardy! appearance of Watson. Noted that structured KB approach is only accurate if the question is framed really well. Watson is based on a massively parallel probabilistic evidence-based architecture.
The question is broken down, statistically analyzed, a hypothesis generated, and if a probable answer can be generate din under 3 seconds it rings in.
The hardware for Watson was not purpose built, and does run on Linux. So how does this apply to healthcare?
Diagnostic assistance, evidenced based, collaborative healthcare.
The information challenge for physicians has evolved from not enough information to too much. The human cognitive capacity is 5 facts per decision, and we are outpaced by the volume of information available.
We all suffer from a Bias Blind Spot that limits us from seeing our own cognitive blind spots, so having a tool to help us overcome that is invaluable.
DeepQA Medical Pipeline process being developed at IBM. The workflow starts with case analysis, and follows the same general principles. Nuance speech recognition is being applied to the Watson technology to enable a more human interface.
IBM notes that this is a heavy lift to have clinicians adopt the concept of DeepDx. A case example from the NEJM was illustrated. Watson started by sourcing out symptoms and adding them to the diagnostic model. Family history added, patient history added, then medications and side effects added. And lastly lab findings are added and the confidence in a diagnoses rockets.
It's like watching a PBL session take place in a fraction of the time. There are risks in having such a system replace the human cognitive approach, but as a tool to support that process, this is valuable.
- Posted using BlogPress from my iPad