Related Experiment Video
Updated: Jan 13, 2026
07:53
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
2.0K
LDSC: enhancing lung disease diagnosis using a simple 1D-CNN.
Mahsa Abbasi1, Fatemeh Imani2, Ali Ghaffari3,4,5
1Department of Computer Engineering, Ta.C, Islamic Azad University, Tabriz, Iran.
Scientific Reports
|January 6, 2026
Summary
This study introduces a new AI tool for diagnosing lung diseases using sound analysis. The lightweight deep learning model achieves high accuracy, offering a promising alternative to traditional stethoscopes for accessible respiratory diagnostics.
Area of Science:
- Biomedical Engineering
- Artificial Intelligence in Medicine
- Respiratory Medicine
Background:
- Traditional stethoscope auscultation for lung disease detection suffers from low sensitivity, complex acoustics, and reliance on clinician expertise.
- These limitations contribute to diagnostic errors, treatment delays, and accessibility issues, particularly in resource-limited settings.
- There is a critical need for automated, affordable, and portable diagnostic solutions for lung illnesses.
Purpose of the Study:
- To develop and evaluate a lightweight deep learning model for automated lung illness classification.
- To assess the performance of the proposed model using Mel-Frequency Cepstral Coefficients (MFCC) features derived from respiratory sounds.
Main Methods:
- Respiratory sound signals were resampled to 4 kHz, enhanced (noise reduction, time-stretch, pitch-shift), and segmented into 3-second frames.
- Normalized Mel-Frequency Cepstral Coefficients (MFCC) were extracted as features from the processed signals.
- A lightweight two-layer one-dimensional convolutional neural network (1D-CNN), termed LDSC, was designed for lung illness classification.
Main Results:
- The LDSC model achieved 98% accuracy, sensitivity, and specificity on the International Conference on Biomedical and Health Informatics (ICBHI) dataset.
- The model demonstrated even higher performance on the King Abdullah University Hospital (KAUH) dataset, reaching 99% accuracy, sensitivity, and specificity.
- These results indicate robust performance in classifying lung illnesses based on acoustic features.
Conclusions:
- The proposed lightweight 1D-CNN model (LDSC) offers a highly accurate and efficient method for automated lung illness classification.
- The approach using MFCC features shows significant potential for developing inexpensive, portable, and reliable respiratory diagnostic tools.
- This technology could improve diagnostic accuracy and accessibility, especially in underserved healthcare environments.
