In the literature, numerous techniques have been employed to decrease noise in medical image modalities, including X-Ray (XR), Ultrasonic (Us), Computed Tomography (CT), Magnetic Resonance Imaging (MRI), and Positron Emission Tomography (PET). These techniques are organized into two main classes: the Multiple Image (MI) and the Single Image (SI) techniques. In the MI techniques, images usually obtained for the same area scanned from different points of view are used. A single image is used in the entire procedure in the SI techniques. SI denoising techniques can be carried out both in a transform or spatial domain. This paper is concerned with single-image noise reduction techniques because we deal with single medical images. The most well-known spatial domain noise reduction techniques, including Gaussian filter, Kuan filter, Frost filter, Lee filter, Gabor filter, Median filter, Homomorphic filter, Speckle reducing anisotropic diffusion (SRAD), Nonlocal-Means (NL-Means), and Total Variation (TV), are studied. Also, the transform domain noise reduction techniques, including wavelet-based and Curvelet-based techniques, and some hybridization techniques are investigated. Finally, a deep (Convolutional Neural Network) CNN-based denoising model is proposed to eliminate Gaussian and Speckle noises in different medical image modalities. This model utilizes the Batch Normalization (BN) and the ReLU as a basic structure. As a result, it attained a considerable improvement over the traditional techniques. The previously mentioned techniques are evaluated and compared by calculating qualitative visual inspection and quantitative parameters like Peak Signal-to-Noise Ratio (PSNR), Correlation Coefficient (
Medical scans are great tools that help specialists to identify the different abnormalities in the body organs. These scans can detect, diagnose, and treat different diseases. The main used medical scans are Us, XR, CT, PET, and MRI [
As shown in
Median filter [
The Log-Gabor filter [
The major impacts of this research work are:
Presenting a comparative analysis of traditional medical image denoising techniques and the proposed CNN-based denoising model for multi-modal images. It is found that the denoising CNN model has a superior performance in contrast to other denoising models. The denoising CNN can easily handle different medical images with different characteristics and different noise types. Therefore, the CNN-based denoising model can improve performance more than other models for various noise levels.
The rest of this paper is arranged as follows. Section 2 reviews the medical image noise reduction techniques. Section 3 illustrates the suggested CNN-based noise reduction model. Section 4 analyses the outcomes and discusses the obtained various results. Section 5 shows the concluding remarks.
In this paper, we are concerned with single image noise reduction techniques. Single-image noise reduction techniques can be carried out in the spatial or transform domains [
It has an impulse response which is a Gaussian function. It can be represented as [
The impulse response of the Gabor filter [
The complex form is given by:
The real is:
The imaginary is:
It is based on the idea that the filtering will be carried out if the variance in a specific region is low or uniform [
It is an adaptive filter suitable for noise reduction [
It is similar to the Lee filter but with various weighting functions [
The weighting function of the Kuan filter is defined as:
Median filtering is implemented by first arranging all pixels from the neighborhood into a numeral arrangement, and the median of these values is computed, and then the filtered pixels are replaced with the computed median [
Generally, an image can be considered as a 2-D function.
The product of illumination (
In fact, it is shown that the SRAD [
For the non-local means filter [
The filter weights can be represented as:
It was presented in [
Visu-shrink makes use of the global thresholding scheme. It adopts a hard threshold value
Sure-shrink denoises an image by applying a soft threshold on the detail coefficients [
Wiener filter belongs to a category of optimal linear filters. It gives a linear estimate of the image from its noisy version. Therefore, this filter needs information about the noise spectrum and the noise-free image [
The transfer function of the Log-Gabor is given as [
The denoising procedure of the log-Gabor filter [ Multi-scale decomposition of the noisy image. Log-Gabor filtering of all sub-bands except for the approximation band. Soft thresholding of the filtered sub-bands. Inverse Discrete Wavelet Transform (DWT).
One of the weaknesses of the wavelet transform is that it is poor at extracting features from curves and edges of images, unlike curvelet transforms. The curvelet transform [
The hybrid filter combines the advantages of a fourth-order PDE and a comfortable intermediate filter [
It is unsuitable for representing the high-level dimension singularities. On the other hand, the curvelet transform is used because it is robust when dealing with image edges, lines, and curves. So, combining both wavelet and curvelet transforms is superior to dealing with noisy images, unlike wavelet only or curvelet only.
In the field of removing noise from images, deep learning structures have been presented for their high quality compared to traditional algorithms. This paper proposes deep learning models to reduce noise from medical images. Deep learning is characterized by its high efficiency; more data is used in the training phase. The proposed model for noise reduction consists of deep residual layers with BN. The remaining layers are distinguished by their ability to differentiate between real features and noise-generated features, and BN was used to achieve stability and speed up the training process. Our proposed model is considered a modification to the model in [
The proposed noise reduction model combines deep residual learning with BN, as shown in
Without deep residual learning, the input density and the convolutional feature are correlated with neighboring ones. Without BN, the problem of internal variable transformation aggravates.
Simulation results are presented using MATLAB R2019a on a Dell machine, Core i5 processor, 8 Gbytes RAMs, and 320 Gbytes hard disk. The metrics [
The higher the PSNR and
System complexity is expressed in terms of processing (CPU) time (seconds), which is calculated from the beginning of the simulation program to the end. The lower the CPU time of the image, the lower the algorithm complexity.
Five examples of scanned images were used. Also, different variances (0.2, 0.1, 0.05, 0.01) of speckle-noise are applied to simulate different scenarios. More than one type of filter has been applied to compare performance.
Technique | Speckle noise variance | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
0.01 | 0.03 | 0.05 | 0.1 | 0.2 | ||||||
PSNR | PSNR | PSNR | PSNR | PSNR | ||||||
28.0102 | 0.9769 | 23.2781 | 0.9353 | 21.0769 | 0.8985 | 18.1543 | 0.8233 | 15.2911 | 0.7128 | |
31.5229 | 0.9895 | 27.0105 | 0.9707 | 24.8439 | 0.9527 | 21.9652 | 0.9116 | 19.0750 | 0.8436 | |
31.8548 | 0.9902 | 28.2434 | 0.9779 | 26.2418 | 0.9648 | 23.4612 | 0.9375 | 20.6344 | 0.8865 | |
30.7178 | 0.9872 | 26.1609 | 0.9642 | 24.1151 | 0.9436 | 21.3156 | 0.9003 | 18.8733 | 0.8378 | |
31.0251 | 0.9936 | 30.2782 | 0.9884 | 29.7176 | 0.9829 | 28.3999 | 0.9698 | 25.9330 | 0.9435 | |
33.6981 | 0.9890 | 31.0804 | 0.9873 | 29.4923 | 0.9853 | 26.8426 | 0.9799 | 24.2598 | 0.9677 | |
29.6997 | 0.9842 | 28.5180 | 0.9780 | 27.4657 | 0.9734 | 25.6790 | 0.9602 | 23.5323 | 0.9332 | |
31.8046 | 0.9932 | 29.9794 | 0.9875 | 28.5259 | 0.9815 | 25.8998 | 0.9657 | 23.0439 | 0.9279 | |
31.2738 | 0.9884 | 29.8405 | 0.9844 | 28.7373 | 0.9811 | 27.3290 | 0.9734 | 25.5154 | 0.9563 | |
30.0209 | 0.9844 | 27.7644 | 0.9714 | 26.3253 | 0.9590 | 24.0820 | 0.9319 | 21.8351 | 0.8829 | |
30.7519 | 0.9872 | 27.7519 | 0.9745 | 26.0192 | 0.9629 | 23.3996 | 0.9359 | 20.8668 | 0.8861 | |
31.2268 | 0.9879 | 28.0981 | 0.9722 | 25.9501 | 0.9583 | 23.6430 | 0.9257 | 20.9773 | 0.8703 | |
33.7201 | 0.9943 | 30.2237 | 0.9872 | 27.0184 | 0.9734 | 22.4560 | 0.9283 | 18.3488 | 0.8308 | |
31.1878 | 0.9885 | 28.8623 | 0.9796 | 27.8430 | 0.9736 | 26.3960 | 0.9631 | 24.4310 | 0.9447 | |
31.4813 | 0.9894 | 29.6291 | 0.9834 | 26.8369 | 0.9680 | 22.1490 | 0.9142 | 17.8571 | 0.8068 | |
30.5038 | 0.9868 | 28.6299 | 0.9797 | 27.7866 | 0.9751 | 26.6413 | 0.9678 | 25.6245 | 0.9588 | |
29.5551 | 0.9836 | 28.6877 | 0.9800 | 28.1031 | 0.9770 | 26.8038 | 0.9688 | 25.1571 | 0.9544 | |
31.1325 | 0.9884 | 28.7915 | 0.9799 | 27.7608 | 0.9737 | 26.0562 | 0.9611 | 24.7046 | 0.9471 | |
34.1454 | 0.9942 | 31.7398 | 0.9898 | 30.5050 | 0.9865 | 28.8336 | 0.9801 | 26.9512 | 0.9695 |
Technique | Gaussian noise variance | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
0.01 | 0.03 | 0.05 | 0.1 | 0.2 | ||||||
PSNR | PSNR | PSNR | PSNR | PSNR | ||||||
20.8417 | 0.9307 | 16.3059 | 0.8180 | 14.2466 | 0.7335 | 11.7785 | 0.5963 | 9.7049 | 0.4527 | |
24.2116 | 0.9686 | 19.6733 | 0.9098 | 17.6424 | 0.8571 | 15.1068 | 0.7558 | 12.9636 | 0.6200 | |
27.4966 | 0.9856 | 23.5744 | 0.9617 | 21.4257 | 0.9380 | 18.6829 | 0.8865 | 15.7750 | 0.7942 | |
22.7385 | 0.9600 | 18.5825 | 0.8886 | 16.6564 | 0.8315 | 14.1818 | 0.7232 | 12.1790 | 0.5930 | |
27.5990 | 0.9888 | 23.5932 | 0.9702 | 21.5189 | 0.9518 | 18.8085 | 0.9089 | 16.5610 | 0.8345 | |
28.1974 | 0.9916 | 24.3441 | 0.9804 | 22.1030 | 0.9684 | 19.1931 | 0.9349 | 16.0889 | 0.8680 | |
25.6345 | 0.9811 | 23.0037 | 0.9623 | 21.0362 | 0.9447 | 18.7088 | 0.8999 | 16.4209 | 0.8253 | |
24.8801 | 0.9852 | 23.4499 | 0.9644 | 21.3680 | 0.9383 | 18.0231 | 0.8764 | 15.6528 | 0.7831 | |
29.4553 | 0.9907 | 25.4594 | 0.9793 | 23.4119 | 0.9673 | 20.4498 | 0.9425 | 17.3240 | 0.8915 | |
27.7727 | 0.9823 | 23.4479 | 0.9472 | 21.0306 | 0.9133 | 18.6600 | 0.8438 | 16.8419 | 0.7400 | |
26.5541 | 0.9797 | 21.3295 | 0.9422 | 19.3624 | 0.9078 | 16.7756 | 0.8311 | 14.4832 | 0.7246 | |
26.6102 | 0.9798 | 21.6901 | 0.9420 | 19.6384 | 0.9051 | 16.8749 | 0.8315 | 14.6064 | 0.7172 | |
24.6479 | 0.9739 | 18.6893 | 0.8963 | 16.3092 | 0.8215 | 13.2565 | 0.6841 | 10.9619 | 0.5287 | |
30.1469 | 0.9948 | 25.5695 | 0.9899 | 23.1681 | 0.9868 | 19.7396 | 0.9806 | 16.3888 | 0.9709 | |
29.5036 | 0.9928 | 26.8895 | 0.9280 | 24.3828 | 0.8486 | 19.5614 | 0.6927 | 13.9213 | 0.5253 | |
32.9572 | 0.9956 | 31.0518 | 0.9931 | 29.6708 | 0.9905 | 27.9753 | 0.9859 | 26.5795 | 0.9805 | |
31.8790 | 0.9943 | 29.2743 | 0.9895 | 27.6129 | 0.9846 | 25.1477 | 0.9730 | 22.6010 | 0.9514 | |
29.9947 | 0.9947 | 25.6157 | 0.9905 | 22.8663 | 0.9867 | 19.6081 | 0.9799 | 16.3728 | 0.9719 | |
31.5927 | 0.9959 | 27.2442 | 0.9922 | 24.8484 | 0.9875 | 21.4395 | 0.9723 | 18.0977 | 0.9238 |
Technique | Gaussian noise variance | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
0.01 | 0.03 | 0.05 | 0.1 | 0.2 | ||||||
PSNR | PSNR | PSNR | PSNR | PSNR | ||||||
21.2504 | 0.9610 | 16.6562 | 0.8887 | 14.6084 | 0.8264 | 12.0746 | 0.7097 | 9.8429 | 0.5654 | |
24.3508 | 0.9825 | 19.8114 | 0.9477 | 17.7356 | 0.9142 | 15.1801 | 0.8410 | 12.8017 | 0.7259 | |
26.8152 | 0.9897 | 23.2161 | 0.9749 | 21.4063 | 0.9616 | 18.6989 | 0.9284 | 15.9557 | 0.8708 | |
22.6126 | 0.9768 | 18.5168 | 0.9356 | 16.5910 | 0.8992 | 14.0305 | 0.8179 | 11.7582 | 0.6908 | |
25.7711 | 0.9898 | 22.4128 | 0.9795 | 20.6254 | 0.9680 | 18.1191 | 0.9413 | 15.5315 | 0.8901 | |
24.6677 | 0.9875 | 22.1698 | 0.9806 | 20.5957 | 0.9732 | 18.0859 | 0.9517 | 15.4732 | 0.9050 | |
25.6804 | 0.9894 | 22.3472 | 0.9785 | 20.6589 | 0.9678 | 15.6298 | 0.9407 | 15.6298 | 0.8897 | |
24.7348 | 0.9855 | 21.7801 | 0.9725 | 19.7765 | 0.9580 | 17.2855 | 0.9295 | 15.1339 | 0.8853 | |
25.7372 | 0.9846 | 23.0762 | 0.9759 | 21.4010 | 0.9690 | 19.3322 | 0.9536 | 16.8377 | 0.9213 | |
26.3055 | 0.9850 | 22.3146 | 0.9628 | 20.5255 | 0.9426 | 18.1173 | 0.8956 | 16.3270 | 0.8150 | |
24.9479 | 0.9781 | 20.5717 | 0.9540 | 18.6896 | 0.9326 | 16.2814 | 0.8856 | 13.9942 | 0.8025 | |
26.7553 | 0.9794 | 21.5293 | 0.9571 | 19.5878 | 0.9356 | 16.5129 | 0.8842 | 14.1402 | 0.7997 | |
24.0124 | 0.9820 | 18.7267 | 0.9338 | 16.3183 | 0.8843 | 13.5036 | 0.7800 | 11.0457 | 0.6388 | |
26.9839 | 0.9922 | 23.2736 | 0.9859 | 21.4941 | 0.9818 | 18.6673 | 0.9750 | 15.8751 | 0.9663 | |
26.8949 | 0.9925 | 25.5056 | 0.9579 | 23.7056 | 0.9098 | 19.6991 | 0.7945 | 14.3280 | 0.6450 | |
26.8886 | 0.9896 | 25.2115 | 0.9848 | 24.3905 | 0.9814 | 23.1479 | 0.9755 | 21.8237 | 0.9665 | |
26.3163 | 0.9881 | 25.0711 | 0.9843 | 24.2884 | 0.9814 | 22.7781 | 0.9734 | 20.8652 | 0.9579 | |
27.2471 | 0.9921 | 23.2134 | 0.9863 | 21.4237 | 0.9821 | 18.7692 | 0.9748 | 15.6992 | 0.9654 | |
27.5218 | 0.9939 | 24.0469 | 0.9890 | 22.1856 | 0.9852 | 19.5071 | 0.9742 | 16.5437 | 0.9440 |
Technique | Gaussian noise variance | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
0.01 | 0.03 | 0.05 | 0.1 | 0.2 | ||||||
PSNR | PSNR | PSNR | PSNR | PSNR | ||||||
21.9919 | 0.9513 | 16.2779 | 0.8687 | 14.3088 | 0.8009 | 11.8568 | 0.6809 | 9.7758 | 0.5347 | |
24.2541 | 0.9783 | 19.7425 | 0.9379 | 17.7246 | 0.9011 | 15.1753 | 0.8207 | 12.9744 | 0.6985 | |
27.5250 | 0.9894 | 23.2695 | 0.9721 | 21.3381 | 0.9571 | 18.5522 | 0.9178 | 15.7534 | 0.8547 | |
22.9991 | 0.9724 | 18.5260 | 0.9230 | 16.7427 | 0.8804 | 14.4317 | 0.7904 | 12.2406 | 0.6585 | |
27.6146 | 0.9915 | 23.7542 | 0.9790 | 21.5806 | 0.9668 | 19.0184 | 0.9377 | 16.3794 | 0.8814 | |
26.3703 | 0.9936 | 24.1481 | 0.9864 | 22.0482 | 0.9785 | 19.0347 | 0.9570 | 16.3938 | 0.9102 | |
27.7304 | 0.9917 | 23.5680 | 0.9792 | 21.7622 | 0.9675 | 19.0638 | 0.9368 | 16.4871 | 0.8815 | |
25.7552 | 0.9901 | 23.5884 | 0.9756 | 21.3679 | 0.9590 | 18.0643 | 0.9176 | 15.0424 | 0.8429 | |
28.8860 | 0.9927 | 24.7521 | 0.9848 | 22.9498 | 0.9771 | 20.2938 | 0.9587 | 17.4668 | 0.9267 | |
26.3212 | 0.9842 | 22.3812 | 0.9605 | 20.7778 | 0.9372 | 18.5784 | 0.8885 | 16.7572 | 0.8021 | |
25.2018 | 0.9820 | 21.2506 | 0.9576 | 19.3351 | 0.9346 | 16.7560 | 0.8812 | 14.4884 | 0.7877 | |
27.8742 | 0.9829 | 22.1955 | 0.9579 | 19.8713 | 0.9326 | 17.0686 | 0.8793 | 14.7805 | 0.7866 | |
24.8477 | 0.9824 | 19.1599 | 0.9339 | 16.6160 | 0.8830 | 13.5993 | 0.7690 | 11.1118 | 0.6212 | |
29.6287 | 0.9956 | 25.2134 | 0.9912 | 23.2785 | 0.9879 | 19.1151 | 0.9826 | 16.1441 | 0.9728 | |
28.1281 | 0.9845 | 26.1022 | 0.9592 | 23.9174 | 0.9362 | 19.4972 | 0.8847 | 14.2592 | 0.8114 | |
28.7748 | 0.9923 | 26.5786 | 0.9872 | 25.3225 | 0.9828 | 23.9894 | 0.9767 | 22.4979 | 0.9665 | |
28.9717 | 0.9926 | 27.2100 | 0.9888 | 26.0103 | 0.9853 | 23.7560 | 0.9748 | 21.6449 | 0.9588 | |
29.8560 | 0.9956 | 25.0852 | 0.9913 | 23.8629 | 0.9883 | 19.3602 | 0.9826 | 16.4108 | 0.9733 | |
29.4463 | 0.9948 | 25.8480 | 0.9900 | 23.7743 | 0.9862 | 20.7508 | 0.9746 | 17.6526 | 0.9402 |
Technique | Gaussian noise variance | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
0.01 | 0.03 | 0.05 | 0.1 | 0.2 | ||||||
PSNR | PSNR | PSNR | PSNR | PSNR | ||||||
20.6595 | 0.9418 | 16.1954 | 0.8458 | 14.2391 | 0.7743 | 11.7852 | 0.6445 | 9.7503 | 0.4978 | |
24.0539 | 0.9721 | 19.7521 | 0.9237 | 17.7470 | 0.8802 | 15.2431 | 0.7896 | 13.0355 | 0.6590 | |
24.9326 | 0.9894 | 21.8559 | 0.9542 | 20.2686 | 0.9333 | 17.7958 | 0.8868 | 15.2203 | 0.8121 | |
22.8040 | 0.9643 | 18.7130 | 0.9083 | 16.9729 | 0.8592 | 14.4578 | 0.7593 | 12.3079 | 0.6278 | |
25.1162 | 0.9800 | 22.5147 | 0.9647 | 20.9914 | 0.9485 | 18.7231 | 0.9098 | 16.4600 | 0.8445 | |
22.9041 | 0.9696 | 21.5898 | 0.9614 | 20.5773 | 0.9518 | 18.5130 | 0.9238 | 16.2626 | 0.8694 | |
24.2906 | 0.9748 | 22.1072 | 0.9602 | 20.6297 | 0.9432 | 18.5023 | 0.9056 | 16.2856 | 0.8343 | |
23.3967 | 0.9726 | 20.5315 | 0.9533 | 18.7582 | 0.9353 | 16.3224 | 0.8987 | 14.2355 | 0.8303 | |
23.9474 | 0.9644 | 21.8962 | 0.9474 | 21.0239 | 0.9346 | 18.9276 | 0.9112 | 17.1131 | 0.8720 | |
24.2459 | 0.9674 | 21.5347 | 0.9383 | 20.0618 | 0.9070 | 18.1824 | 0.8468 | 17.0809 | 0.7492 | |
22.9471 | 0.9384 | 20.0592 | 0.9077 | 18.6359 | 0.8830 | 16.3888 | 0.8241 | 14.3448 | 0.7280 | |
25.0336 | 0.9414 | 20.8461 | 0.9131 | 19.4042 | 0.8871 | 16.7860 | 0.8277 | 14.5672 | 0.7256 | |
23.7400 | 0.9724 | 18.5731 | 0.9092 | 16.1815 | 0.8463 | 13.4955 | 0.7289 | 11.1892 | 0.5755 | |
27.2613 | 0.9877 | 23.8999 | 0.9785 | 22.1224 | 0.9716 | 19.4694 | 0.9603 | 16.1088 | 0.9426 | |
27.1016 | 0.9778 | 25.5965 | 0.9457 | 23.6533 | 0.9162 | 19.4450 | 0.8544 | 14.1395 | 0.7703 | |
23.9425 | 0.9723 | 22.1255 | 0.9575 | 21.2340 | 0.9472 | 20.1232 | 0.9316 | 18.8790 | 0.9061 | |
22.5104 | 0.9609 | 21.6525 | 0.9522 | 21.0674 | 0.9445 | 20.0320 | 0.9284 | 18.8220 | 0.9040 | |
27.2223 | 0.9876 | 24.2374 | 0.9777 | 22.2588 | 0.9711 | 19.3878 | 0.9606 | 17.9452 | 0.9418 | |
28.0531 | 0.9895 | 24.6735 | 0.9790 | 22.8408 | 0.9700 | 20.2205 | 0.9499 | 18.4544 | 0.9317 |
Visual results of the Us breast image, shown in
From the presented results, the efficiency of the traditional filters is low when the noise level increases, as shown by the results of
The CNN model for Gaussian noise reduction excels from the MRI image in
Technique | Image processing CPU time (s) | ||||
---|---|---|---|---|---|
Us | XR | CT | PET | MR | |
0.05 | 0.03 | 0.03 | 0.03 | 0.03 | |
0.16 | 0.05 | 0.06 | 0.05 | 0.06 | |
1.2 | 0.7 | 0.68 | 0.7 | 0.65 | |
0.03 | 0.02 | 0.02 | 0.02 | 0.02 | |
6.9 | 3.4 | 3.4 | 3.4 | 3.5 | |
12.3 | 9.3 | 8.9 | 9.1 | 8.9 | |
0.14 | 0.13 | 0.12 | 0.11 | 0.12 | |
0.56 | 0.13 | 0.12 | 0.11 | 0.11 | |
0.12 | 0.11 | 0.09 | 0.09 | 0.12 | |
1.6 | 0.5 | 0.5 | 0.4 | 0.5 | |
0.69 | 0.3 | 0.2 | 0.3 | 0.3 | |
13.3 | 7.3 | 7.1 | 7.1 | 6.9 | |
33.4 | 23.3 | 23.8 | 23.5 | 23.2 | |
6.4 | 3.4 | 1.92 | 1.41 | 1.03 | |
8.3 | 7.4 | 3.2 | 4.4 | 1.9 | |
19.7 | 17.5 | 8.2 | 7.8 | 6.7 | |
34.1 | 25.4 | 25.1 | 25.6 | 25.3 | |
7.07 | 4.58 | 4.03 | 4.45 | 2.78 |
The obtained results demonstrate that the Gaussian and Gabor filters cannot remove Speckle or Gaussian noise, especially with large noise variances. The Gaussian filter also smoothes images and blurs the edges. The median filter is more robust than the Gaussian filter because it preserves edges. The Lee filter has a smoothing effect if the area has low variance. However, it fails to remove noise from areas closer to edges and lines. The Frost filter relies on adaptive filtering between pixels to reduce noise and smoothes the homogeneous regions. The Kuan filter has the advantage of preserving sharp edges compared to Frost and Lee filters. However, it is not effective at high noise variances.
The Homomorphic filter can maintain the brightness of images and increase the contrast. In the SRAD model, its performance decreases with increasing noise variance. However, it can remove speckle noise while maintaining edge features. In investigating various image denoising methods based on wavelet transform, it has been found that the sub-band adaptive thresholding methods outperformed the highest spatial domain method in the MRI image for all noise levels. If they did not perform better, they were slightly better than the Gaussian and Gabor filters. Out of the sub-band adaptive methods, sure-shrink consistently outperformed and visushrink. Because sure-shrink filtering is adaptive when dealing with images that contain abrupt changes or boundaries.
On the other hand, visushrink removes too many coefficients and overly smooths images. Wiener filter yields better results when the image is corrupted with a Gaussian noise rather than speckle noise. The Log-Gabor filter is poor when dealing with noisy images. The curvelet denoising technique is suitable when dealing with image edges and curves. The NL-mean filter is characterized by its ability to preserve the clarity of images and less information loss, especially at low noise variance. This is the result of taking the average of all pixels in the image as opposed to local mean filtering. The TV algorithms have advantages such as mean filtering and linear dimming that preserve edge characteristics and reduce noise in flat areas, at most high and low noise variances.
Hybrid algorithms are characterized by combining the features of discrete algorithms such as Hybrid Model 1, which outperforms other algorithms in terms of edges and structures. Hybrid Model 2 was also implemented by combining wavelet and curve transformation models. The curvelet transform model is characterized by representing curves in the images, and the wavelet transform model is characterized by reducing noise in smooth areas. One of the problems with the Hybrid Model 2 is that it takes a longer time to process. However, it significantly outperforms wavelet-based technologies.
In contrast to traditional noise reduction techniques, the CNN noise reduction model can handle noisy images with different noise levels. It also reduces quantitative and qualitative noise. The CNN model is characterized by its ability to adapt in all modalities of medical images, unlike other models that excel in some modalities and fail in other modalities of images. Moreover, visual comparisons of the different algorithms show that the CNN noise reduction model produces more perceptive images with sharp edges and finer information.
This paper presented a CNN-based denoising model and a comparative study of noise reduction methods for various medical image modalities. The performance of all employed methods is tested on different medical images (Us, X-ray, CT, PET, and MR). This study can summarize all the noticed advantages and disadvantages of the tested noise reduction techniques employed in this paper on different medical image modalities. The employed algorithms are tested on the medical images, and their denoising performance has been compared and studied. For the spatial domain, median filtering outperformed the Gaussian filtering. The superiority of the waveform model in reducing Gaussian noise was proven to be significantly superior to the rest of the other models. It has shown the best image quality. From comparing the denoising results of different denoising techniques using different threshold functions, it is obvious that the images become blurred after global threshold denoising. On the other hand, the image texture details are well preserved using the wavelet denoising method based on adaptive thresholding. The comparison is more obvious in the MRI image, which has more texture details. Similar results can be obtained for other test images. The curvelet algorithms excel at reducing noise in some image modalities such as curvilinear images and linear singularities. Moreover, the hybrid algorithm based on combining wavelet and curvelet transforms maximizes the advantages of both. The numerical and visual results show that the proposed CNN noise reduction model is adapted to all modalities of medical images in terms of perceptual and visual quality. It also possesses high scores in most noise levels and image modalities. Therefore, we conclude that the CNN model is superior to traditional filtering and noise reduction techniques. The CNN model also has a good denoising CPU time. For future work, combinations of other transforms and the adaptation of CNN models could yield better results than those obtained separately.
Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2022R66), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia.