Monday, November 26, 2012

RNSA 2012 - Decision Support in Clinical Practice

Decision support systems use knowledge -- ranging from books, to web sites, to real-time artificial intelligence systems -- to help physicians improve their decision making.
Passive decision support leav responsibility to search the data with the radiologist while active decision support uses artificial intelligence and computer software to actively provide relevant information and guidance to the radiologist.
In the US a lot of the work in decision support in order entry is stimulated by the appropriateness criteria for radiological examinations.
Decision support systems have to be real time and integrated to be able to be a success in clinical practice.
Recipe for success of decision support depends on the system being:
- available
- easy to use
- integrated
- collaborative
- good medicine
- aware of limits

Predicting Diagnosis and Outcome
Currently there is a clearly identified role for informatics in predicting diagnosis and outcome based on variables derived from imaging. However, there are important trade-offs that exist when developing or using predictive models.
The goal of decision suppprt is to take predictive information and assist the physician in quantifying risk of disease or probability.
An important trade-off is that In most cases the correctness of the outcome of decision support heavily depends on the dataset used to construct or train the decision support system.

Quantitative Image Analysis for Image Retrieval, Decision Support, and Knowledge Discovery
Quantitative image analysis is to characterize images and or parts of images with rich features that can be accessed by computers for comparison or decision support.
Image characterization can be done by annotation and computing of image features. This characterization (among others based on quantitative image analysis) can be used for content based image retrieval by defining a vector of different features. The acknowledged features in this vector should be weighted to determine their relative importance. This weighting can either be done by human definition or by computer learning or training and depends on what you are interested in. In radiology this can be used to retrieve similar images from the PACS or, more challenging, find images with similar lesions. Standardization using e.g. RadLex and identifying exact locations is vital to obtain usefull desciptions to use as image features.

Key points
1) Evolving technologies provide new ways to integrate advanced decision support into routine clinical practice, and decision support systems can improve outcomes in patient care.
2) Decision support gives radiology a chance to transform to a more proactive role in the managing of patients and examinations.

No comments:

Post a Comment