Source-linked AI summary
Exponential Pixelating Integral transform with dual fractal features for enhanced chest X-ray abnormality detection
Naveenraj Kamalakannan, Sri Ram Macharla, M Kanimozhi, M S Sudhakar
TL;DR
Respiratory-disorder diagnosis from chest X-rays is hindered by low contrast, overlapping anatomy, noise, and interpretive variability. The paper introduces EPI-enhanced fractal representations classified with pairwise MARS ensembles, achieving high performance across respiratory-disease categories while retaining a mathematically interpretable design.
Problem
Chest X-ray diagnosis is challenging because respiratory disorders share anatomical patterns and images contain subtle intensity differences, overlapping structures, and noise.
Method
The method applies EPI contrast enhancement, converts images into polar Mandelbrot and Julia representations, and classifies the combined features with ensembled MARS.
Results
EPIFF-MARS exceeded 97% across reported metrics, including 99.12% accuracy, 99.48% specificity, and an averaged F1 score above 98%.
Takeaways & Limitations
The framework provides an automated and interpretable approach for distinguishing healthy and diseased patients across multiple respiratory-disorder categories.
Takeaways & Limitations
Deep-learning comparators may achieve slightly higher accuracy, while the Julia representation assumes an initial z value of zero and iterates under a fixed threshold.
Abstract
from arXiv · showhide
The heightened prevalence of respiratory disorders, particularly exacerbated by a significant upswing in fatalities due to the novel coronavirus, underscores the critical need for early detection and timely intervention. This imperative is paramount, possessing the potential to profoundly impact and safeguard numerous lives. Medically, chest radiography stands out as an essential and economically viable medical imaging approach for diagnosing and assessing the severity of diverse Respiratory Disorders. However, their detection in Chest X-Rays is a cumbersome task even for well-trained radiologists owing to low contrast issues, overlapping of the tissue structures, subjective variability, and the presence of noise. To address these issues, a novel analytical model termed Exponential Pixelating Integral is introduced for the automatic detection of infections in Chest X-Rays in this work. Initially, the presented Exponential Pixelating Integral enhances the pixel intensities to overcome the low-contrast issues that are then polar-transformed followed by their representation using the locally invariant Mandelbrot and Julia fractal geometries for effective distinction of structural features. The collated features labeled Exponential Pixelating Integral with dually characterized fractal features are then classified by the non-parametric multivariate adaptive regression splines to establish an ensemble model between each pair of classes for effective diagnosis of diverse diseases. Rigorous analysis of the proposed classification framework on large medical benchmarked datasets showcases its superiority over its peers by registering a higher classification accuracy and F1 scores ranging from 98.46 to 99.45% and 96.53-98.10% respectively, making it a precise and interpretable automated system for diagnosing respiratory disorders.
1. Introduction
Respiratory disorders are difficult to diagnose from chest radiographs because overlapping anatomy, subtle intensity differences, and variable expertise complicate interpretation. The paper motivates an automated approach using EPI, fractal feature representations, and MARS classification, evaluated on large CXR datasets.
- Clinical motivation: Early identification of respiratory disorders matters because prompt diagnosis and intervention can slow disease progression, reduce symptoms, and decrease exacerbations.
- Clinical motivation: Chest radiography is fast, less sensitive, and more cost-effective than CT, making it widely used for respiratory disease identification, especially in low-resource hospitals.
- Clinical motivation: Chest radiographs are difficult to interpret because superimposed anatomy, subtle lesions, and limited expertise can produce inaccurate and time-consuming assessments.These challenges motivate automated diagnostic techniques to support radiologists.
- Existing approaches: Prior analytical and machine-learning approaches include wavelet-based nearest neighbors, fuzzy clustering, entropy-fractal features, and ensembles, but reported methods have limitations such as moderate performance or poor localization.
- Proposed approach: The proposed framework uses Exponential Pixelating Integral for low-contrast enhancement, Mandelbrot and Julia geometries for texture representation, and ensembled MARS for robust classification.
- Evaluation: The system was evaluated on 100,000 chest radiographs from Kaggle, RSNA, and NIH datasets, with results indicating improved classification performance after combining EPI and fractal geometry.
2. Methodology
The methodology enhances subtle CXR intensity differences with EPI, then combines polar, Mandelbrot, and Julia representations to extract structural features. These features support classification through a framework using EPI-derived enhancement and dual fractal characterization.
- 2.1 Exponential Pixelating Integral (EPI) model: EPI computes exponential cumulative intensity behavior within local 3×3 neighborhoods to enhance low-contrast details while emphasizing subtle lesion-related variations.The transformed values are normalized and contrast-stretched to reduce illumination effects, highlight regions of interest, and improve discrimination.
- 2.1 Exponential Pixelating Integral (EPI) model: The input image is resized and divided into non-overlapping subregions before localized EPI features are extracted.The stated dimensions are m×n for the image and r×r for the localized subregions, with r=3 specified for the kernel.
- 2.1 Exponential Pixelating Integral (EPI) model: Fig. 2 shows original, normalized EPI-transformed, and contrast-stretched images, with EPI emphasizing subtle intensity variations and lesion-related regions.The transformed images are described as making specific image features easier to identify and analyze.
- 2.2 Fractal Representation: The EPI image is polar-transformed, then represented with Mandelbrot and Julia fractal geometries to capture complex shapes and structural detail.The approach uses polar coordinates for representing symmetrical forms, while the Julia representation uses a fixed complex additive and image-dependent values.
- 2.2.2 Julia Set Image Representation: The Julia representation differs from the Mandelbrot representation by fixing the complex additive and focusing visualization on a specific value rather than the full range of possible values.The resulting Julia structures contain clusters and boundaries whose intensity variation is used for analyzing the mapped source image.
3. Performance Analysis
EPIFF-MARS was evaluated on Chest X-ray 14 and COVID-19 datasets using confusion-matrix measures, classification metrics, ROC analysis, and comparisons with recent methods.
- Datasets: The model was investigated on Chest X-ray 14 and COVID-19 Radiography datasets containing 112,120 and 21,165 CXRs, respectively.Chest X-ray 14 images were frontal-view 1024×1024 grayscale images, while the COVID-19 dataset contained 256×256 PNG images and masks.
- Confusion-matrix analysis: True-positive and true-negative rates exceeded 98% and 97%, while false-positive and false-negative rates remained below 1% and 3% across multiple respiratory-disorder categories.The categories included Emphysema, COVID-19, Edema, Pneumonia, and Effusion, despite data imbalance.
- Classification metrics: EPIFF-MARS exceeded 97% across the metrics in Table 4, with averaged sensitivity of 97.72% and specificity of 99.48%.The reported evaluation included accuracy, precision, sensitivity, specificity, and F1 score.
- Model interpretation: Pairwise MARS classification and distinctive fractal features were associated with improved handling of imbalanced data and nuanced class boundaries.The analysis attributes this to selecting essential category-representing features and differentiating specific class pairs.
- Comparative performance: 99.12% accuracy and 97.87% F1 score were reported for EPIFF, alongside 98.56% AUC, 98.11% precision, and 97.64% recall, surpassing listed peers.The comparison used existing state-of-the-art schemes, including Red Deer, XG-Boost-Beta–T, ACPL, and HealthyGAN.
- ROC analysis: ROC analysis showed AUC values consistently above 0.98 for all classes, indicating discrimination between abnormal and non-abnormal CXRs.The ROC analysis plotted true-positive rate and false-positive rate, with performance assessed through the area under the curve.
EPIFF-MARS
EPIFF-MARS combines EPI contrast enhancement, polar transformation, Mandelbrot and Julia fractal features, and MARS classification for respiratory-disorder detection. Across benchmark, ablation, robustness, and scalability analyses, the pipeline improves diagnostic performance and maintains efficiency under noise and feature reduction.
- EPIFF-MARS: 99.10% accuracy was achieved for COVID-19 detection, with strong ROC performance and lower complexity than several deep-learning competitors.The model narrowly trailed CheXNet on F1 and MKSC on precision and recall, while offering lower computational demands.
- EPIFF-MARS: 99% average accuracy was achieved with 40% feature reduction, while processing remained under 20–30 ms.MARS selected disease-specific subsets of relevant features as the maximum feature dimension varied.
- EPIFF-MARS: Over 99% average accuracy and specificity, with average F1 scores and sensitivity above 98%, were obtained on a balanced 1,500-image dataset.The balanced subset contained 300 images for each disease category.
- EPIFF-MARS: 7% higher accuracy, 21% higher F1 score, 23% higher sensitivity, and 4% higher specificity resulted from adding fractal features to EPI-plus-polar features.Raw features reached 78% average accuracy and 49% F1, whereas EPI features reached 83.2% accuracy and 60.1% F1 before fractal augmentation.
- EPIFF-MARS: Consistent ROC output across Gaussian, salt-and-pepper, and speckle noise was attributed to EPI’s smoothing effect.The transformation averages noise and reduces its visibility in the image representations.
4. Computational complexity
The computational-complexity analysis decomposes EPIFF-MARS into image transformation, fractal generation, and MARS stages. Its time cost includes pixel processing plus basis-function and feature terms, with processing times compared against peer models.
- 4. Computational complexity: The total time complexity is expressed as O(n × m) + O(B + (N × F)).The formulation combines image-sized processing with MARS-related basis-function and feature terms.
- 4. Computational complexity: EPI processing an n × m image incurs O(n × m) time complexity.The analysis treats the EPI stage as pixel-wise processing.
- 4. Computational complexity: Processing times were compared with peer models using the relative-analysis results in Table 14.The comparison evaluates practical computational efficiency across models.
EPIFF-MARS 57
EPIFF-MARS is presented as computationally simple and suitable for implementation on constrained hardware. Its efficiency is attributed to less intricate developmental stages and reduced resource requirements.
- EPIFF-MARS 57: EPIFF-MARS has exceptionally short processing time compared with peer models, reflecting high simplicity.The authors attribute this efficiency to less intricate developmental stages.
- EPIFF-MARS 57: The model’s space requirement includes O(n × m) image and fractal matrices plus MARS storage for basis functions and features.The space analysis identifies image-sized storage for transformations and additional storage associated with MARS.
- EPIFF-MARS 57: EPIFF-MARS is described as easy to implement on simple hardware with low computational cost for resource-constrained settings.This conclusion follows from the reported time and space-complexity analysis.
5. Conclusion
The conclusion presents EPI as a local exponential moving-average transform whose polar Mandelbrot and Julia representations support MARS-based respiratory-disorder classification. Benchmark, noise, and complexity analyses are reported as evidence of consistent performance and practical usability.
- 5. Conclusion: EPI computes an exponential moving average over overlapping local 3 × 3 pixel kernels to standardize grayscale intensities and highlight structures.The transform is intended to address small differential intensity values in chest radiographs.
- 5. Conclusion: Polar Mandelbrot and Julia representations enhance differentiation between normal tissue structures and abnormalities before pairwise MARS classification.Features from the three intermediaries are collated into an ensemble model for respiratory-disorder classification.
- 5. Conclusion: Benchmark, noise, and computational-complexity analyses reported consistent and superior EPIFF-MARS performance over peer methods.The conclusion characterizes the approach as robust and practical based on these evaluations.