Utilizing Machine Learning and 3D Neuroimaging to Predict Hearing Loss - A Comparative Analysis of Dimensionality Reduction and Regression Techniques - Proposed machine learning framework for age-related hearing loss prediction

Aug 4, 2026

Upload by Asher

Project research develops machine learning methods to predict hearing loss thresholds from gray-matter 3D brain images. The workflow is built in two phases: first, 3D CNN-based representation learning with autoencoders/variational autoencoders for dimensionality reduction and feature-rich decoding; second, regression using the learned latent or reconstructed features. Experiments compare random forest, XGBoost, and multilayer perceptron, evaluating performance primarily with RMSE. Results show the lowest RMSE with multi-layer perceptron and test-set ranges of 8.80 (PT500) and 22.57 (PT4000), supporting more accurate, data-driven diagnostics.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages5

Reading Time13 min

AccessFree to download

Introduction

Literature Review

Methodology

Data Acquisition

Data Preparation

Utilizing Machine Learning and 3D Neuroimaging to Predict Hearing Loss: A Comparative Analysis of Dimensionality Reduction and Regression Techniques Trinath Sai Subhash Reddy Pittala School of Computing Clemson University [tpittal@g.clemson.edu](tpittal@g.clemson.edu) Uma Maheswara R Meleti School of Computing Clemson University [umeleti@g.clemson.edu](umeleti@g.clemson.edu) Manasa Thatipamula School of Computing Clemson University [mthatip@g.clemson.edu](mthatip@g.clemson.edu) arXiv :2405 .00142v2 [ cs .LG] 2 May 2024 Abstract—In this project, we have explored machine learning approaches for predicting hearing loss thresholds on the brain’s gray matter 3D images. We have solved the problem statement in two phases. In the first phase, we used a 3D CNN model to reduce high-dimensional input into latent space and decode it into an original image to represent the input in rich feature space. In the second phase, we utilized this model to reduce input into rich features and used these features to train standard machine learning models for predicting hearing thresholds. We have experimented with autoencoders and variational autoencoders in the first phase for dimensionality reduction and explored random forest, XGBoost and multilayer perceptron for regressing the thresholds. We split the given data set into training and testing sets and achieved an 8.80 range and 22.57 range for PT500 and PT4000 on the test set, respectively. We got the lowest RMSE using multi-layer perceptron among the other models. Our approach leverages the unique capabilities of VAEs to capture complex, non-linear relationships within highdimensional neuroimaging data. We rigorously evaluated the models using various metrics, focusing on the root mean squared error (RMSE). The results highlight the efficacy of the multi-layer neural network model, which outperformed other techniques in terms of accuracy. This project advances the application of data mining in medical diagnostics and enhances our understanding of age-related hearing loss through innovative machine-learning frameworks. 1. Introduction Age-related hearing loss (ARHL), also known as presbycusis, is an increasingly prevalent condition that affects a significant portion of the aging population. Characterized by the gradual loss of hearing capabilities, ARHL poses substantial challenges to interpersonal communication and overall quality of life. Traditional methods of diagnosing and evaluating ARHL rely on auditory tests and essential medical imaging; however, these methods often fall short of identifying the underlying neuroanatomical changes associated with the condition. Recent medical imaging and machine learning advancements present new opportunities to enhance our under- standing of ARHL. Specifically, brain magnetic resonance imaging (MRI) offers detailed insights into the brain’s structural integrity. At the same time, sophisticated data mining techniques enable the analysis of complex datasets to reveal patterns not immediately apparent to human observers. This project leverages these technologies to bridge the gap between neuroanatomical changes and auditory function, employing a dataset that combines gray matter images with auditory threshold measurements. Our research employs advanced machine learning techniques, including variational autoencoders (VAEs) and multi-layer neural networks, to analyze the relationship between brain structure and hearing loss. These models are particularly suited for this task because they can handle high-dimensional data and learn deep representations that capture the subtle nuances of brain morphology related to ARHL. By training these models on a robust dataset from clinical settings, this project aims to predict hearing thresholds more accurately and identify specific brain regions that correlate with hearing decline. Through this approach, the project seeks to contribute to the broader field of medical diagnostics by providing a m

FAQs

What two phases does the project use to predict hearing loss thresholds?

The first phase learns and compresses 3D gray-matter images into latent representations using autoencoders/VAEs, then decodes them to rich feature space. The second phase trains regression models on the learned features to predict hearing thresholds.

Which regression models are compared for threshold prediction?

The study evaluates random forest, XGBoost, and multilayer perceptron for regressing the 500 Hz and 4000 Hz hearing thresholds.

How is model performance evaluated in the project?

Performance is evaluated using multiple metrics with a focus on root mean squared error (RMSE), comparing how accurately each model predicts the thresholds on a held-out test set.

Aug 4, 2026

Upload by Asher