Estimating personalized diagnostic rules depending on individualized characteristics

Abstract

There is an increasing demand for personalization of disease screening based on assessment of patient risk and other characteristics. For example, in breast cancer screening, advanced imaging technologies have made it possible to move away from “one-size-fits-all” screening guidelines to targeted risk-based screening for those who are in need. Since diagnostic performance of various imaging modalities may vary across subjects, applying the most accurate modality to the patients who would benefit the most requires personalized strategy. To address these needs, we propose novel machine learning methods to estimate personalized diagnostic rules for medical screening or diagnosis by maximizing a weighted combination of sensitivity and specificity across subgroups of subjects. We first develop methods that can be applied when competing modalities or screening strategies are observed on the same subject (paired design). Next, we present methods for studies where not all subjects receive both modalities (unpaired design). We study theoretical properties including consistency and risk bound of the personalized diagnostic rules, and conduct simulation studies to examine performance of the proposed methods. Lastly, we analyze data collected from a brain imaging study of Parkinson’s disease using positron emission tomography (PET) and diffusion tensor imaging (DTI) with paired and unpaired designs. Our results show that in some cases a personalized modality assignment is estimated to improve empirical AUC compared to a “one-size-fits-all” assignment strategy.

Machine learning applications using diffusion tensor imaging of human brain: A PubMed literature review

“…All the nine studies related to movement disorders involved patients with Parkinson's disease (PD). Classification of Parkinson's patients versus healthy controls was performed in three studies [133,134,135] with some differences in the reference standard for the disease (sTable 7 in our supplementary document). One study by Liu et al [136] involved the identification of a biomarker through feature selection using the United Kingdom brain bank criteria for diagnosis.…”

“…Of the eight studies that performed classification, SVMs were used in seven [133,134,137,138,139,140,141] and LR was used in one study [135]. One study [136] performed feature selection and used Folded concave penalized model for the same.…”

Nonparametric estimation of linear personalized diagnostics rules via efficient grid algorithm

“…2,3 Our goal is different in that we focus on subgroup-specific biomarkers for detecting diseases at the earliest possible stage before treatments or even any clinical symptoms begin. The closest relevant work by Liu et al 4 identified subgroup-specific biomarkers by maximizing the absolute mean difference of the biomarkers. However, their distance-based criteria was less commonly used for evaluating biomarkers in practice.…”