Factor Parameter Ranking for Prostate Prognosis Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for predicting the pathological malignancy of prostate cancer require invasive examinations, are time-consuming, and costly, and face challenges in efficiently narrowing down combinations of parameter candidates for accurate prognosis estimation due to high computational demands and large candidate sets.
Innovation Solution
A suggestion device that learns estimation and rating models to identify high-ranking candidates among parameter combinations by using machine learning and rating systems like Elo Rating, reducing calculation load and time by selectively competing and ranking candidates based on performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all candidate parameter combinations are evaluated using machine learning estimation models, then the accuracy of prognosis prediction is improved, but the calculation amount and calculation time become enormous
Solution Approach 1:
The patent segments the evaluation process into two distinct phases: a calculation phase where estimation models are trained on candidate parameter combinations, and an evaluation phase where a pre-learned rating model quickly assesses remaining candidates. This segmentation allows computationally intensive model training to be performed once, followed by rapid evaluation of all candidates using the rating model, thereby reducing total calculation time while maintaining prediction accuracy.
Solution Approach 2:
The patent performs preliminary action by learning the rating model in advance using data from the calculation phase. The rating model is trained on the relationship between parameter combinations and prediction performance, enabling it to quickly evaluate candidates without requiring full estimation model training. This preliminary preparation of the rating model allows for rapid candidate evaluation while preserving the ability to achieve high prediction accuracy.
2Measurement precision
If the number of parameter combinations is increased to improve prediction accuracy, then the estimation performance is improved, but the number of candidates causes an explosion in competition number
Solution Approach 1:
The patent introduces a rating model as an intermediary between the estimation models and the candidate evaluation process. The rating model learns from the estimation models during the calculation phase and then serves as a mediator to quickly evaluate all candidate parameter combinations during the evaluation phase. This intermediary enables the system to handle a large number of candidates efficiently without requiring exhaustive computation for each candidate, thus managing the explosion in competition number while maintaining estimation performance.
3Measurement precision
If more examinations are performed to acquire more parameter data, then the estimation accuracy is improved, but the cost, effort, and time required increase
Solution Approach 1:
The patent applies parameter changes by systematically varying the parameter combinations across different candidates to identify which specific parameters and their combinations yield the best estimation accuracy. Rather than uniformly performing all possible examinations, the system evaluates different parameter subsets and identifies the optimal combination, thereby achieving high estimation accuracy while minimizing the number of examinations required and reducing associated costs and efforts.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure suggests a factor parameter for estimating a label of a target from a plurality of parameters that can be acquired for the target. In a suggestion device (101), a calculator (102) learns, by referring to a plurality of records in which each record represents a plurality of parameter values acquired from the target and a label provided to the target, an estimation model that estimates a label from a parameter related to a combination for a part of calculation candidates among a plurality of candidates representing a combination of parameters; calculates performance of the estimation model; causes calculation candidates to compete based on the calculated performance; and calculates a rating value of the calculation candidate. An estimator (103) learns a rating model by a rating value and a combination of parameters of the calculation candidate, and estimates a rating value of an estimation candidate other than the calculation candidate. An outputter (104) outputs, as a useful candidate representing a factor parameter, a candidate having a calculated or estimated high-ranking rating value and being the calculation candidate.