A brain tumor is a malignant disease with a high mortality ratio, occurring when one or more tissues in the brain become malignant and start growing unusually. Moreover, it disturbs the brain’s normal function and other neighboring tissues. The detection of tumors depends on the size and position of the malignant tissue.

The diagnosis of brain tumors is established with a neurological examination of brain tissues [1]. The examination is based on the careful observation of muscle movement and strength, a reflexive and radiological examination of the brain tissues. These examinations are performed with the use of medical imaging methods. These techniques include the computed tomography scan (CT), magnetic resonance imaging technique (MRI), and positron emission tomography (PET) [2]. The growth and pressure of growing tissue also damage vital tissues of the brain that also need to be detected by medical imaging. One of the main issue types is water accumulation in the brain that is caused due to the tumor and results in the circulation of cerebrospinal fluid. The growth of tumors also increases symptoms. The primary symptoms are headaches, hearing problems, loss of smell, skull swelling, and sensational loss of vision problems. Moreover, it can also cause insomnia, memory disorders, and speaking power reduction in the patient.

Magnetic resonance imaging (MRI) is a noninvasive examination to diagnose the different types of brain tumors. These scans use attractive fields and radio frequencies rather than X-rays and computers to create complete pictures of the brain’s soft tissue. A 3-dimensional visual slice of the brain is taken from the coronal axial and sagittal directions in the MRI test. A contrast dye is used to improve the scanned image [11].

Positron emission tomography (PET) is valid for diagnosing primary brain tumors. A PET scan detects changes in cells as they grow. Radioactive glucose is injected in a minimal amount compared with standard brain parts. Tumor cells absorb a specified amount of radioactive glucose depending on the tumor’s grade. A PET scan can help determine the grade and severity of a tumor [12].

Functional magnetic resonance imaging (fMRI) is a method for determining which portion of the brain is responsible for a specific function [13,14]. FMRI can focus on brain activity changes in regions as small as one cubic millimeter using the most powerful magnets. Because it is noninvasive, a significant advantage of MRI is that it can be readily used on human subjects. FMRI allows patients to perform routine tasks during the scan. Unlike X-rays, computed tomography (CT), and positron emission tomography (PET) scans, fMRI does not involve radiation exposure. If appropriately performed, fMRI presents no dangers. It is safe, noninvasive, and effective in assessing brain function [15].

Magnetic resonance spectroscopy (MRS) is another noninvasive technology that can detect chemical abnormalities in body tissues. It can assess disorders such as brain issues, multiple sclerosis, Alzheimer’s disease, tumors, and scarring after radiation. It shows some points of the brain tumors that are not clear on MRI scans. MRS is like MRI, and the only difference is that it shows function instead of structure [16].

In some cases, diagnosis cannot happen through scans. A biopsy is performed to conclude the type of tumor [17]. A biopsy is a surgical process to eradicate a small tumor for examination. A tiny tissue sample is taken and examined under the microscope by a doctor. A biopsy can be performed as a separate spinal biopsy process with a small hole drilled in the skull [18].

Diffusion tensor imaging (DTI) is an MRI method that employs anisotropic diffusion to determine the brain’s axonal (white-matter) architecture. Brain pathways may be analyzed with fiber tractography (FT), a 3D reconstruction approach that uses data from DTI. Diffusion tensor imaging (DTI) is used to estimate the location and anisotropy of the brain’s white-matter tracts. This approach gives a detailed image of the brain’s structure and is used to assess changes over time. This scan allows for surgical treatment by visualizing the brain’s circuitry for surgery [19].

Anat. Levin et al. reported a colorization approach using an optimization technique to color available grayscale images. This approach is automated and does not need operator involvement. In this YUV color space used for applying colors, they colorized images and movie clips. They considered pixels to be a neighborhood that was timewise adjacent to space. This approach did not require precise segmentation and user interaction [20]. Jacob and Gupta proposed a semiautomated color-transfer approach to convert grayscale images and videos into colored versions. The proposed method was applied to still images and videos and compared with the automated segmentation method. This method requires human interaction, where a user indicates the desired “color marker” to each region. However, the automated segmentation method did not detect exact boundaries and tedious tasks [21]. Dhaniya et al. used pivoting, editing, zooming, and CLACHE evening out to maximize the available data. The use of k implies bunching, brain tumors were separated, and component foci were gathered. Support vector machine classifier (SVM) image order computation aids in accurately recognizing a tumor in its early stages. Their work demonstrated how to combine essential k implies grouping with dropout arrangement and SVM information contention to create a new viewpoint in the field of picture order, which was extensively studied [22].

Blasi and Reforgiato proposed a technique used for colorization by transferring color from a source image to an image [23]. The proposed method was inspired by the algorithm of Welsh et al. and worked based on the similarity image enhancing technique. The critical feature of this new technique was the adaption of efficient data and large data structures to retrieve “color words” from a vast vocabulary. This approach does not work on nonhomogeneous images [24]. T. Tan et al. discussed the colorization of CT brain images. They introduced a technique based on a histogram to enhance the visualization of brain tumors to minimize human error. The pseudo-coloring method was used to color the abnormal regions of a brain, such as a tumor or an injury. The technique improved accuracy and time by 13.3% and 21.6%, respectively [25]. A. A. shah et al. proposed a colorization approach to grayscale medical images using user interaction to convert grayscale medical images into colored images and then convert the colored images again into grayscale images. Their presented method was applied to 40 X-ray, dental, and MRI images. This methodology does not work on CT images [26]. Attique et al. discussed the two methods to increase the discrimination between normal and abnormal tissues using CIELAB color and space. The second method automatically segmented the three regions of a T2 brain MR image by using the approach of auto centroid selection. The auto centroid segmentation method efficiently separates gray and white matter, and cerebrospinal fluid (CSF). The proposed method achieved better segment results when compared with watershed and Gaussian classifier segmentation methods [27]. Lagodzinski and Smolka used morphological distance transformation to assign colors using scribbling techniques on different medical images.

Colors were automatically propagated in the different regions of brain MR images by following standard distance transform, hybrid distance transform, and color blending steps. The key feature was the preservation of the intensity of the original image. However, this method was unsuitable for coloring the available images [28]. Zhao Yuanmeng et al. first selected an appropriate chromatic image for reference to colorize images. The grayscale and colored (reference) image was converted into a lαβ color space. Then, the best matching chromatic pixel for each pixel of the grayscale image was selected after this chromatic value had been transformed from the best matching reference pixel to the corresponding grayscale pixel. This is a complicated technique and not valuable in coloring available tumor images [29]. Popowicz and Smolka used the isoline concept of a geographical map and a distance map between grayscale pixels and the colored regions scribbled by the user. The method is efficient because accuracy is maintained without losing any information. The accuracy of the proposed algorithm is 100%, but the drawback of this approach is that a reference image is required [30].

Martinez-Escobar et al. suggested three steps: colorization, segmentation, and postprocessing, with a region growing technique to segment the tumor region. This method for coloring medical images is beneficial and supportive [31]. Yogesh Rathore et al. proposed a fully automated approach. The steps in this approach are creating an image database, converting the reference images into a lab color space, creating a histogram and signature, and saving the images with a signature in the database. This approach achieves better results, and the quality may be improved in the future by using more color transfer techniques [32]. Z. Cheng et al. discussed a fully computerized approach with a perfect patch match technique to color grayscale images without user interaction. The proposed method uses a neural network to train on a large number of images to remove the burden from the user. It requires a vast database, which is possible for authentic images, but impossible for synthetic images. This method is better than traditional methods [33]. Michal Kawulok et al. reported a colorization approach for scribble images using textural features instead of luminance and gradient values. Textural features were accessed using linear discriminative analysis (LDA). This technique works well but achieves poor results for region boundaries due to the large kernel size [34]. Li B. et al. presented a technique based on deep learning to extract engraved areas from a stela’s 3D scanned mesh. First, the mesh’s unequal distribution of vertices was modified using a mesh subdivision method, resulting in a mesh with evenly distributed vertices. The subdivided mesh was then used to extract surface characteristics (depth, concave features, and local surface features). In experiments, the authors demonstrated that the suggested approach efficiently removes engraved portions of inscriptions from a stela’s rough surface. It is resilient to noisy and severely abraded characters. The suggested technique of the authors surpassed the other best method by 2.95%, 3.65%, and 7.53% in terms of F1 score, IoU, and SIRI, respectively [35].

Kawulok et al. suggested a colorization approach based on an extreme example picture and used a new locality consistent sparse representation. Their system automatically colorizes the target grayscale image by sparse pursuit after providing a single reference color image. The approach acts at the superpixel level for efficiency and durability. Each superpixel’s descriptor is composed of minor intensity characteristics, midlevel texture features, and significant semantic features, which are then concatenated. Their experimental results showed that aesthetically and numerically, their colorization approach surpassed the latest methods [36]. R.jayadevan et al. presented several pseudo-coloring techniques to convert grayscale images into colored versions. User interactive (ICT), semiautomated (CTGI, PHI), and fully automated (CTTS, CGFA, FDPC) pseudo-coloring techniques were discussed, and comparisons were drawn among them. Results showed that PHI was better than the CTGI method (semiautomated). PDPC achieved excellent and superior results to CTTS and CGFA (fully automated) [37]. Liu and Zhang proposed an automated colorization approach. However, a source image is required to provide color information on the luminance histogram of grayscale and color images. Then, the zero point is found to calculate the average color from the source image. The mapping between luminance color and target image generates the colorization results. The purposed method works well on both normal and noisy images, but a large number of texture parameters are required to perform colorization [38]. A. Bugeau and V.T. Ta reported a patch base colorization approach using reference images and a distance selection framework for color prediction. This method converts from the RGB to the YUV color space because lαβ gives visually inconsistent results. Color is transferred to grayscale images using luminance features. Then, the distance selection framework and color prediction are performed, and total variation on the output images is applied to smoothen the colorization results [39].