Evaluating the Impact of Image Enhancement Techniques on Deep Learning-Based X-ray Classification

The research evaluates different image enhancement approaches regarding their impact on deep learning algorithms which detect body regions in X-ray scans. We analyze how Bilateral Filtering as well as Contrast Limited Adaptive Histogram Equalization (CLAHE) and Wavelet Denoising and Super-Resolution influence X-ray image quality which subsequently impacts Convolutional Neural Networks (CNNs) classification results. The evaluation demonstrates Bilateral Filtering delivers superior performance than other enhancement processes according to PSNR and SSIM evaluations on LEG, CTScan and Chest X-ray datasets. The experimental results for the LEG dataset demonstrated Bilateral Filtering produced a higher PSNR of 51.78 along with an SSIM of 0.99918 compared to CLAHE which resulted in inferior PSNR of 19.55 and SSIM of 0.67681. Results from Wavelet Denoising and Super-Resolution matched those of Bilateral Filtering with PSNR values at 44.65 and SSIM values at 0.99559. The evaluation of combined enhancement techniques with CNN-based classification resulted in perfect test set accuracy at 100%. This proves that both methods produce highly accurate results when integrated together. This research increases the general understanding of preprocessing methods which work best for medical imaging and classification procedures.