This repo presents a Python implementation of the Adaptation-Classification Framework introduced by
Kitamoto, T., Idé, T., Tezuka, Y. et al., "Identifying primary aldosteronism patients who require adrenal venous sampling: a multi-center study," Scientific Reports 13, 21722 (2023) [link].
The primary goal of this framework is to identify primary aldosteronism patients who could benefit from specific surgical treatment.
As the name implies, the proposed framework comprises two modules:
- Data adaptation module,
- Patient classification module.
The key assumption is that it operates in a multicenter setting. In other words, it utilizes a well-established reference dataset from one medical institution to build these models and employs a form of transfer learning to apply the models to data collected at other medical institutions. The overall problem setting is explained here.
For domain adaptation with a limited number of samples, a new algorithm called bpca_impute has been developed. This module implements a data imputation approach using Bayesian probabilistic principal component analysis. One significant advantage of our BPCA-based imputation is that it is essentially parameter-free. In impute_bpca_ard, the dimensionality of the latent principal subspace, which is the critical parameter in any PCA-based algorithm, is automatically determined through an automatic relevance determination (ARD) mechanism. This feature makes it a preferred choice when dealing with a limited number of samples.
For more technical details, please refer to the notebooks:
For the technical detail, see the notebooks: