Information-theoretic view of the scheduling problem in whole-body computer aided detection/diagnosis (CAD)

摘要

A method for automatically scheduling tasks in whole-body computer aided detection/diagnosis (CAD), including: (a) receiving a plurality of tasks to be executed by a whole-body CAD system; (b) identifying a task to be executed, wherein the task to be executed has an expected information gain that is greater than that of each of the other tasks; (c) executing the task with the greatest expected information gain and removing the executed task from further analysis; and (d) repeating steps (b) and (c) for the remaining tasks.

主权项

1. A method for automatically scheduling tasks in whole-body computer aided detection/diagnosis (CAD), comprising:
receiving a plurality of statistically dependent tasks to be executed by a whole-body CAD system; actively scheduling the tasks by:
(a) identifying a task to be executed, wherein the task to be executed has an expected information gain that is greater than that of the other remaining tasks, wherein the expected information gain is represented byIGy=∑i⁢(H⁡(xi❘y∈Ψ)-∫y∈Ψ⁢H⁡(xi❘y∈Φ)⁢p⁡(y)⁢⁢ⅆy), wherein H(xi|y∈Ψ) is a conditional entropy before executing the task and H(xi|y∈Φ) is a conditional entropy after executing the task, xi is a goal of the task, y is an outcome after executing the task, Ψ is a distribution of the outcome before executing the task and Φ is a distribution of the outcome after executing the task;
(b) executing the task with the greatest expected information gain and removing the executed task from further analysis; and(c) repeating steps (a) and (b) for the remaining tasks.