View of Image Characterization Based Fetal Brain MRI Localization and Extraction

12  Download (0)

Full text


Image Characterization Based Fetal Brain MRI Localization and Extraction

P.Durgadevi1, S.Vijayalakshmi2

1 (Research Scholar (Ph.D.), School of Computing Science and Engineering, Galgotias University, Greater Noida, Uttar Pradesh, [email protected])

2 (Professor, School of Computing Science and Engineering, Galgotias University, Greater Noida, Uttar Pradesh, Tamil Nadu,[email protected])


Most fetal cerebrum MRI reproduction calculations depend just on mind tissue-pertinent voxels of low-resolution (LR) pictures to upgrade the nature of between cut movement adjustment and picture remaking. Therefore the fetal cerebrum should be restricted and removed as an initial step, which is typically a relentless and tedious automatic or self-loader task. The planned work to utilize epoch-coordinated format pictures as earlier information to automatize cerebrum restriction and extraction. This has been accomplished through a novel programmed cerebrum restriction and extraction strategy dependent on vigorous format to- cut square coordinating and deformable cut-to-layout enlistment. The layout based methodology has likewise empowered the remaking of fetal mind pictures in standard radiological anatomical planes in a typical to arrange space. The coordinated this methodology into the new recreation pipeline that includes power standardization, between cut movement remedy, and super-resolution (SR) remaking. To this end a novel methodology dependent on the projection of each cut of the LR cerebrum veils into the format space utilizing a combination procedure. This has empowered the refinement of cerebrum covers in the LR pictures at each movement remedy emphasis. The general mind restriction and extraction calculation have appeared to deliver cerebrum veils that are near physically drawn cerebrum covers, demonstrating a normal Dice cover proportion of 94.5 percentages. It likewise exhibited that receiving a cut to-format enrolment and proliferation of the mind cover .cut by-cut prompts a huge improvement in cerebrum extraction execution contrasted with worldwide unbending cerebrum extraction and thusly like the tidal recreated pictures.

Appraisals performed by two master onlookers show that the proposed pipeline can accomplish comparative recreation quality to reference remaking dependent on automatic cut by-cut cerebrum extraction. In this paper, initially characterize the images and then segment the localized image by using seeded region growing segmentation techniques automatically and results were discussed.

Keywords: Fetal MRI, Segmentation, automatic, head localization, seeded region growing segmentation


Fetal MRI has pulled in a ton of consideration and is by and large gradually utilized as a corresponding symptomatic device to pre-birth ultrasound imaging as it gives a superior delicate tissue contrast. Quick single shot multi-cut MRI successions are utilized to freeze maternal and fetal movement; however, the securing of thick cuts (around 2 to 4 mm) is important to get adequate Signal-to-Noise Ratio (SNR) given the short procurement time


used to stay away from the movement at each cut obtaining. Thick cuts and between cut movement antiques limit the precision of volumetric investigation for clinical determination and neuroscience considers [1].

In the most recent years, interest in finding a high-goal (HR) volumetric picture given a bunch of low-goal stacks made out of thick cuts with between cut movement antiques has developed extensively [2]. In, the principal reproduction procedures dependent on cut to- volume enrolment and dissipated information addition was presented [3]. Afterward, super- goal (SR) strategies [4]. It supported the nature of the reconstructed picture by displaying an opposite issue for fetal picture recreation [5]. By giving better subtleties of the fetal cerebrum, such methods have empowered the neuroscience local area [6] to perform new exploration on early human mental health. The fetal MRI remaking pipeline comprises of different picture preparing steps [7] (force normalization, movement assessment, and SR reproduction). When all is said in done, calculations depend just on cerebrum tissue- significant voxels of low-resolution (LR) pictures to warrant the suspicion of movement inflexibility utilized in unbending movement remedy [8]. This is an urgent advance of the reproduction calculations. Therefore, the fetal mind (or cerebrum area) should be first confined and afterward separated before movement assessment and SR recreation [9]. These two successive cycles are known as (1) cerebrum confinement and (2) mind extraction.

Mind restriction normally plans to identify a jumping enclose containing the cerebrum the obtained pictures. Fetal mind extraction, then again, intends to portray (cover) the cerebrum in the obtained pictures [10].

Fetal mind limitation and extraction is commonly done physically or semi-naturally, consequently compares to the most tedious, non-programmed step of the whole picture reproduction pipeline [11]. It is thusly not a sensible answer for huge scope contemplates. In the writing, even though precise mind extraction instruments have been created for grown- up and newborn child cerebrum MRI those apparatuses are not promptly pertinent to fetal MRI. Fetal mind MRI varies from multiple points of view from neonatal or grown-up cerebrum MRI regarding picture content (with maternal tissues encompassing the fetal cerebrum), picture contrast, mind size, and particularly the subjective (non-standard) fetal position and direction which likewise changes because of movement. On-going examinations have tended to the issue of mind confinement as well as extraction in fetal MRI by receiving either format based segmentation [12].

In this work, present format based brain localization, extraction, and division in a standard direction and basic organize space for an in-iitero fetal cerebrum MRI [13]. In particular, our contributions are:

 The Automated cut by-cut mind extraction technique in each heap of thick cuts (LR picture). It couples strong layout to-cut square coordinating dependent on La standard improvement for programmed mind confinement with a novel deformable slice-to- template brain extraction technique.

 The mix of the veiling cycle with power standardization, movement revision, and super-goal recreation: mind covers are refined in the spatial space of the format utilizing an agreement combination casting a ballot cycle and are re-applied to the LR pictures as the reproduction continues.


 A broad approval of our mind extraction and restriction techniques on clinical information including sound and neurotic situations where:

It shows the impact of the pivot inspecting step boundary of our cerebrum limitation calculation on the mind recognition achievement rate.

Analyze the achievement rate and normal run season of our cerebrum restriction method against one of the cutting edge mama chine-learning strategies.

It assesses the mind extraction execution as far as to cover measures with automatic depictions

The effect of programmed cerebrum extraction execution on the last reproduction quality,

It directs a perceptual assessment by master spectators to think about the nature of the last remaking utilizing mind covers acquired physically and consequently with the proposed technique.

 It performs mind volumetric dependent on the reproduced pictures: intra-cranial fetal cerebrum volume is registered straightforwardly from our cerebrum extraction and the misuse of the basic reference space to perform mind division. Fig. 1 shows the samples of automatic MRI localization and extraction.

Fig. 1. Illustration images of MRI localization and extraction II. OUTLINE OF FETAL BRAIN LOCALIZATION AND EXTRACTION

A few, moderately later, contemplates have tended to the programmed confinement as well as extraction of the fetal cerebrum in MRI through either format based division [14]. AI the main endeavor [15] of fetal mind extraction proposed to assess the area of the eyes (because of unbending format enrolment) to section the fetal cerebrum utilizing contrast, morphological and biometrical earlier data [16]. This strategy gave exact mind depictions in 22 out of 24 MRI piles of embryos matured somewhere in the range of 30 and 35 gestational weeks; be that as it may, they depended on the presumption of minor movement between cuts, which restricts the power of the technique to clinical information bases. A managed approach, in light of a two-stage irregular woodland classifier, was received in [17] to get a technique pertinent to every fetal age and more hearty as for (w.r.t) movement between cuts.

This technique has indicated practically identical outcomes to the strategy yet the entire mind was contained inside the last bouncing box in just 28°« (coronal) to 58percentage (cross- over, sagittal) of the cases. Afterward, limitation exactness of the mind was radically improved by consolidating earlier information on the fetal head size with maximally stable external locales discovery, packaged Scale-Invariant Feature Transform (SIFT) highlights, and a sack of-words model: the entire cerebrum was contained inside the last jumping


enclose 85percentage of the cases [18]. The strategy, restricted to the limitation of the fetal cerebrum (bounding box), was additionally improved with the utilization of circular Gabor descriptors and 2D level-sets to give a precise last division of the fetal mind with a Dice cover metric over ninety percentage [19].

All the more as of late, those methods have been coordinated in the recreation pipeline.

Mind restriction, done physically by reorienting and trimming the LR pictures as an initial step, and programmed multi-map book based cerebrum extraction have been joined with movement correction and SR reproduction [20]. In any case, such a automatic limitation forestalls a mechanized picture reconstruction pipeline. The creators have proposed to limit the fetal mind utilizing a Bag-of-Words model utilizing SIFT highlights in addition to the random sample consensus (RANSAC) technique for hearty fitting and to give a division of the cerebrum utilizing a blend of an arbitrary woodland classifier and a 3D restrictive irregular field. Mind extraction is then refined as the recreation advances, producing the last division of the reproduced fetal cerebrum with a ninety-three percentage of mean dice value.


The proposed technique has a few phases. In the initial step, the fetal head is localized by image characteristics such as entropy, standard deviation (SD), mean, contrast, texture, intensity, tissues, skewness, kurtosis, covariance, and variance, and so on. In the excess stages, the characterized fetal head is handled and the extraction of fetal MRI is done by using the seeded region growing segmentation method. The following algorithms describe the steps of the proposed algorithm.

 Initially feed the input brain image

 Slice the given brain images

 Characterize the slice image based on its features pixel values

 Convert the sliced image into the binary image

 Segment the processed brain image by using the seeded growing segmentation method

 After segmenting the selected brain image, it will terminate the process.

1. Input Used

The materials utilized for this work are gotten from The input images are having 617 X617 pixels. These images are used to characterize to find the values of mean, contrast, intensity, tissues, skewness, kurtosis, covariance, and variance.

2. Localization of fetal MRI

The confinement of fetal head in two-dimensional (2D) fetal MRI segmentation [22].

The fetal cerebrum is the biggest organ in the embryo, and the more obvious element in MRI information is the low force of the fetal skull and focused energy of cerebrospinal liquid (CSF). Thus, this essential cycle was done on characterizing the image. The fundamental challenge is that the baby is in an obscure position, encircled by maternal tissues. In any MRI that attention on imaging the fetal head will fundamentally have the cerebrum consuming the essential space in the picture. In these pictures, confining cerebrum parcel in MRI can be accomplished by finding the Pinion [21] of the picture. Restricting the fetal MRI


segmentation consequently takes out the maternal tissue [23]. The input fetal images are exposed in Fig. 2.

Fig. 2. Input fetal image

The Mean esteem gives the commitment of individual pixel force for the whole picture and change is regularly used to discover how every pixel shifts from the adjoining pixel (or focus pixel) and is utilized to characterize various districts.

The mean value T of the fetal images is calculated as:

𝑇 = 𝑚𝐼=1 𝑛𝑗 =1𝐴(𝑖, 𝑗)

𝑚𝑥𝑛 (1) Where,

An (i, j) is defined as the intensity’s pixels at ith &jth location.

The standard deviation of the picture suggests that picture is variable. The standard deviation is a proportion of this variability. A picture is an assortment of information focuses on the light force, deviation of the picture suggests a gross proportion of the imprecision/variety about the objective estimation of light power, at each such information point. Low change or standard deviation implies the pixel forces are nearer to the mean and high fluctuation implies the pixel powers are a long way from the mean. Contrast is high for pictures with exclusive expectation deviation esteems.

The standard deviation D of the fetal images is computed as:

𝑆𝐷 = 1

𝑁 (𝑥𝑖− 𝑦



) (2) Wherever,

N is the number of samples of fetal images

𝑦 = 𝑥𝑖 (3)




y is the mean and S is the standard deviation 𝑥𝑖 𝑖𝑠 𝑡𝑕𝑒 𝑖𝑡𝑕 𝑝𝑖𝑥𝑒𝑙,

𝑁 = 𝑚𝑥𝑛, 𝑠 𝑡𝑕𝑒 𝑡𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑖𝑥𝑒𝑙𝑠 𝑎𝑛𝑑 𝑖 = 1, 2, 3, 4, … . 𝑁.

Using the skewness as a proportion of balance. If the skewness of S is zero, at that point the dissemination spoke to by S is symmetric. On the off chance that the skewness is negative, at that point the conveyance is slanted to one side, while if the slant is positive, at that point the dispersion is slanted to one side.

The skewness of fetal images S as follows:


𝑆 = 𝑛 𝑛𝑖=1 𝑥𝑖− 𝑦 3

(𝑛 − 1)(𝑛 − 2)𝑠3 (4)

Kurtosis gives an estimation about the limits (for example brain) of the dissemination of information, and in this manner gives a sign of the presence of exceptions.

The kurtosis of a fetal images R as follows:

𝑅 = 1


𝑥𝑖−µ 𝜎 4

𝑛𝑖=1 (5)

Where µ is the mean of the fetal images and σ is the standard deviation of fetal images.

The variance is a proportion of inconstancy or range within a bunch of information.

Numerically, the normal standard deviation can be calculated from the mean score. It utilizes an accompanying recipe of the process of the variance of fetal images.

The variance of fetal images Var (A) as calculated:

Var (A) = Σ (Ai - B) 2 / N = Σ ai2 / N (6) Wherever,

N: the no. of fetal images in a given set of fetal images B: mean of N fetal images.

Ai: ith fresh fetal images in the given amount of fetal images ai: ith deviance fetal images in the given amount of fetal images Var(X): variance of fetal images of the given sample dataset.

Covariance is a proportion of the degree to which relating components from two arrangements of requested information move a similar way. It utilizes the accompanying equation to figure fetal image covariance.

The covariance of fetal images Cov (X) as calculated:

Cov(X) = Σ (Xi - x) (Yi - y) / N = Σ xiyi / N (7) Where,

N is the no. of fetal images in every group of facts x is the mean of the N scores in the head dataset Xi: ith fresh score in the first set of fetal images

xi is the ith deviation fetal images in the first set of fetal images y is the mean of the N fetal images in the second data set

Yi is the ith raw score in the second set of fetal images Yi is the ith deviation score in the second set of fetal images

Cov(X) is the covariance of corresponding fetal images in the two sets of data.

Entropy is a factual proportion of irregularity that can be used to portray the surface of the info picture. Entropy E of the insignificant hippocampus is characterized as:

𝐸 = 𝑠𝑢𝑚 𝑝𝑋𝑙𝑜𝑔2 𝑝 (8) Where,

In the entropy (E) the p is defined as fetal image counts in the minimal range.

Differentiation is the proportion of the powerful contrast between a pixel and its neighbor over the entire image. Differentiation is the distinction in luminance or shading that makes an item (or its portrayal in a picture or show) recognizable. In the visual impression of this present reality, contrast is dictated by the distinction in the tone and splendor of the article and different items inside a similar field of view. The human visual framework is more delicate to differentiate than outright luminance; they can see the world correspondingly


paying little mind to the gigantic changes in light throughout the day or here and there. The most extreme difference in a picture is the differentiation proportion of dynamic reach.

The contrast C of the insignificant fetal pictures is followed as:

𝐶 = 𝑖 − 𝑗 2 𝐴 𝑖, 𝑗 (9)



An ( i,j) is the strength's pixel at (i, j)th position.

The above-mentioned characteristics features of the images value ranges, minimum and maximum value are calculated automaticly by using sample input pages are shown in Table 1.

In the second stage, the test cuts are taken for the division. For the information cut, the estimations of the highlights E, S, T, Cov (X), R, and σ are registered. These qualities are contrasted and the values are given in Table I. If the processed estimations of E, S, T, Cov (X), R, and σ lie in the reach, at that point it is accepted that the information cut contains the hippocampus and is considered for division, else it is disposed of. The proposed strategy continues with the division measure with the legitimate info cut. The legitimate cut is then upgraded to get clear limits. Improvement helps the division cycle relatively simpler. To improve the picture a middle channel with the local grid is applied [26]. Middle separating is a nonlinear activity frequently used to lessen "salt and pepper" commotion. A middle channel is more successful at the same time lessen commotion, what's more, safeguard edges. Each yield pixel contains the middle worth in the m x n neighborhood around the relating pixel in the information picture. The following stage is to make the fetal images conspicuous by utilizing base cap separating. For this, a level, disk-shaped organizing component (SE) is made, with the area. The localized fetal images are extracted by using the seeded region growing segmentation method.

3. Seeded region growing segmentation method

Region growing is a basic area based picture division technique [24]. This way to deal with division inspects adjoining pixels of beginning seed focuses and decides if the pixel neighbors ought to be added to the district. The cycle is iterated on, in a similar way as broad information bunching algorithms. Image division is the way toward grouping pixels into remarkable picture locales (i.e.) districts relating to singular surfaces, items, or normal pieces of articles. Picture division assumes an imperative part in picture examination and PC vision applications. A few broadly useful calculations and procedures have been created for picture division. Region growing ought to be halted when the locale of the premium is isolated from the information picture. In light of the application, locale of interest may contrast and subsequently, none of the division calculations fulfills the worldwide applications. In this manner, division stays a difficult region for analysts. At last, an automatic seeded growing segmentation technique is proposed for fragmenting shading pictures. Fig. 3 are the samples of the seeded growing segmentation method.


Fig. 3 samples of seeded growing segmentation method.

Seeded Region Growing is a calculation utilized for picture division. Introductory seeds are chosen dependent on a few models and afterward, the division is finished by locale developing cycle. Locale development is done by contrasting the seed pixel and all the adjoining pixels with an edge and pixels which are over the edge are assembled to frame a solitary locale. The area development cycle is done work all the pixel values are gathered in any of the districts. Introductory seeds might be a solitary pixel or gathering of pixels called bunches [26]. In the wake of collection due to the presence of clamor over-segmentation may result and this is stayed away from by district consolidating measure [27].

For then again automatic depiction, altering devices, for example, MATLAB as a rule show 3D information as 3 synchronized 2D symmetrical perspectives (sagittal, coronal, and hub) onto which the administrator draws the form of the objective design. The yield information along these lines comprises a progression of 2D shapes from which a nonstop 3D surface must be extracated. This is a nontrivial postprocessing task and is inclined to mistakes. For example, because of interslice irregularities in the division, knocks in the recreated 3D surface are inescapable. More strong division strategies can normally be gotten from genuine 3D design models in that they can guarantee worldwide smoother and more reasonable surfaces across cuts.

The essential planes of MRI: from top to down (hub plane), from front to back (coronal plane), and side to side (sagittal plane) [28]. In the X-Y-Z organize framework, a hub is an X-Y plane, corresponding to the ground, the head from the feet. A coronal is an X- Z plane, the front from the back. The sagittal, coronal, and axial views of the fetal brain are shown in Fig 4.

Fig .4 Segmented Fetal brain from MRI in sagittal view, coronal view and axial view Dice coefficient (D), a few famous quantitative measures are accessible for contrasting primary likenesses of two pictures. Dice coefficient (D) is an estimation of imbalance data for two twofold areas E and F.

The Dice coefficient [D] is given by:


𝐷 E, F = 2 E ∩ F

E + F (10)

Where E and F are two informational indexes. The qualities D fluctuates from 0 for a complete contradiction to 1 for complete arrangement, among E and F.

The Sensitivity (Sn) is the level of ROI voxels perceived by a calculation and Specificity (Sp) is the level of non-ROI voxels perceived by a calculation.

Sn and Sp are computed as:

𝑆𝑛 = TA

TA +FB (11)

Sp = TB


Trials were done by applying the proposed strategy on the material pool which was imaged at various gestational weeks. These five volumes were divided physically by a specialist clinician and these physically fragmented cerebrum partitions (best quality level) are utilized for looking at and assessing the presentation of the proposed programmed technique. The specialists, by contrasting the outcomes and physically divided outcomes, open that the outcomes acquired from the proposed programmed technique are in acceptable concurrence with automatic outcomes. For subjective evaluation, the fragmented mind parcel by the proposed technique has appeared in Fig. 5.

Fig. 5 Segmented Fetal brain

Fetal Brain Extracted (Ibr) from the original fetal MRI image using cover IM as:

𝐼𝑏𝑟 𝑚. 𝑛 = 𝐼 𝑚. 𝑛 𝐼𝑀 𝑚. 𝑛 = 1

0 𝑜𝑡𝑕𝑒𝑟𝑤𝑖𝑠𝑒 (13)

Fig .6 (a) Original Image (b) diffused image c) filtered image (d) binary image (e) brain mask (f) segmented fetal brain.


Table 1. Features of segmented fetal image Features Value of the

image element

E 1.0828

S 0.0476

T 0.4882

Cov(X) min(0.049) max(0.059)

R 1.0037

Σ 11.2

Fig .6 describes the segmented fetal brain image from the original brain MRI after applying the process of masking the fetal brain MRI image. For quantitative assessment, the qualities for Dice coefficient D, affectability Sn, and particularity Sp are registered for the proposed technique by accepting the physically portioned results as the highest quality level and are given in table 2.

Table 2 Computed values of D, Sn, and Sp for the proposed method.


This article discussed the subtleties of different factual measures concerning computerized picture handling. Introduced a factual model for choosing the factual measure appropriately before going for an intricate picture handling procedure. In the proposed model rules of the yield, necessity is considered while choosing the factual measure. MATLAB recreations are done dependent on different factual measures separating procedure to demonstrate the utilization of each at the root level. Subsequently, it can presume that the proposed factual model is can be utilized as a pre-handling model for different computerized picture preparing strategy to improve the adequacy of complex picture handling procedure in the following levels and results are too discussed by feeding sample dataset to extract the fetal brain, initially characterize the images and then segment the localized image by using seeded region growing segmentation techniques automatically are discussed with the results

Test Images

Dice (D)

Sensitivity (Sn)

Specificity (Sp) Orientations Vol.



1 0.884

0.93 1.10

2 0.94 0.97 1.01

3 0.92 0.93 1.06


1 0.87 0.95 1.07

2 0.95 0.97 1.04


and the fetal brain extraction modeled local fetal minima will be carried out in future the detection of abnormality in embryo brain magnetic resonance imaging automatically.


[1] Gholipour, A., Estroff, J. A., Barnewolt, C. E., Robertson, R. L., Grant, P. E., Gagoski, B., &

Mulkern, R. V. (2014). Fetal MRI: a technical update with educational aspirations. Concepts in Magnetic Resonance Part A, 43(6), 237-266.

[2] Rousseau, F., Glenn, O. A., Iordanova, B., Rodriguez-Carranza, C., Vigneron, D. B., Barkovich, J.

A., & Studholme, C. (2006). Registration-based approach for reconstruction of high-resolution in utero fetal MR brain images. Academic radiology, 13(9), 1072-1081.

[3] Xue, H., Srinivasan, L., Jiang, S., Rutherford, M., Edwards, A. D., Rueckert, D., & Hajnal, J. V.

(2007). Automatic segmentation and reconstruction of the cortex from neonatal MRI. Neuroimage, 38(3), 461-477.

[4] Gholipour, A., Estroff, J. A., & Warfield, S. K. (2010). Robust super-resolution volume reconstruction from slice acquisitions: application to fetal brain MRI. IEEE transactions on medical imaging, 29(10), 1739-1758.

[5] Keraudren, K., Kuklisova-Murgasova, M., Kyriakopoulou, V., Malamateniou, C., Rutherford, M.

A., Kainz, B., ... & Rueckert, D. (2014). Automated fetal brain segmentation from 2D MRI slices for motion correction. NeuroImage, 101, 633-643.

[6] Tourbier, S., Hagmann, P., Cagneaux, M., Guibaud, L., Gorthi, S., Schaer, M., ... & Cuadra, M. B.

(2015, March). Automatic brain extraction in fetal MRI using multi-atlas-based segmentation.

In Medical Imaging 2015: Image Processing (Vol. 9413, p. 94130Y). International Society for Optics and Photonics.

[7] Keraudren, K., Kainz, B., Oktay, O., Kyriakopoulou, V., Rutherford, M., Hajnal, J. V., &

Rueckert, D. (2015, October). Automated localization of fetal organs in MRI using random forests with steerable features. In International conference on medical image computing and computer- assisted Intervention (pp. 620-627). Springer, Cham.

[8] Shishegar, R., Joshi, A. A., Tolcos, M., Walker, D. W., & Johnston, L. A. (2017, April).

Automatic segmentation of fetal brain using diffusion-weighted imaging cues. In 2017 IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017) (pp. 804-807). IEEE.

[9] Ison, M., Donner, R., Dittrich, E., Kasprian, G., Prayer, D., & Langs, G. (2012, October). Fully automated brain extraction and orientation in raw fetal MRI. In Workshop on Paediatric and Perinatal Imaging, MICCAI (pp. 17-24).

[10] Kuklisova-Murgasova, M., Quaghebeur, G., Rutherford, M. A., Hajnal, J. V., & Schnabel, J. A.

(2012). Reconstruction of fetal brain MRI with intensity matching and complete outlier removal. Medical image analysis, 16(8), 1550-1564.

[11] Mehta, R. (2017). Population-specific template construction and brain structure segmentation using deep learning methods (Doctoral dissertation, Ph.D. thesis, International Institute of Information Technology Hyderabad).

[12] Chen, J., Fang, Z., Zhang, G., Ling, L., Li, G., Zhang, H., & Wang, L. (2020). Automatic brain extraction from 3D fetal MR image with a deep learning-based multi-step framework. Computerized Medical Imaging and Graphics, 88, 101848.

[13] Anquez, J., Angelini, E. D., & Bloch, I. (2009, June). Automatic segmentation of head structures on fetal MRI. In 2009 IEEE International Symposium on Biomedical Imaging: From Nano to Macro (pp. 109-112). IEEE.


[14] Al-Faris, A. Q., Ngah, U. K., Isa, N. A. M., & Shuaib, I. L. (2014). Breast MRI tumor segmentation using modified automatic seeded region growing based on particle swarm optimization image clustering. In Soft Computing in Industrial Applications (pp. 49-60). Springer, Cham.

[15] Lou, J., Li, D., Bui, T. D., Zhao, F., Sun, L., Li, G., & Shen, D. (2019, October). Automatic fetal brain extraction using multi-stage u-net with deep supervision. In International Workshop on Machine Learning in Medical Imaging (pp. 592-600). Springer, Cham.

[16] Khalili, N., Lessmann, N., Turk, E., Claessens, N., de Heus, R., Kolk, T., ... & Išgum, I. (2019).

Automatic brain tissue segmentation in fetal MRI using convolutional neural networks. Magnetic resonance imaging, 64, 77-89.

[17] Lou, J., Li, D., Bui, T. D., Zhao, F., Sun, L., Li, G., & Shen, D. (2019, October). Automatic fetal brain extraction using multi-stage u-net with deep supervision. In International Workshop on Machine Learning in Medical Imaging (pp. 592-600). Springer, Cham.

[18] Raja, N. S. M., Fernandes, S. L., Dey, N., Satapathy, S. C., & Rajinikanth, V. (2018). Contrast- enhanced medical MRI evaluation using Tsallis entropy and region growing segmentation. Journal of Ambient Intelligence and Humanized Computing, 1-12.

[19] Kainz, B., Keraudren, K., Kyriakopoulou, V., Rutherford, M., Hajnal, J. V., & Rueckert, D.

(2014, April). Fast fully automatic brain detection in fetal MRI using dense rotation invariant image descriptors. In 2014 IEEE 11th International Symposium on Biomedical Imaging (ISBI) (pp. 1230- 1233). IEEE.

[20] Mostapha, M., & Styner, M. (2019). Role of deep learning in infant brain MRI analysis. Magnetic resonance imaging, 64, 171-189.

[21] Khalili, N., Lessmann, N., Turk, E., Claessens, N., de Heus, R., Kolk, T., ... & Išgum, I. (2019).

Automatic brain tissue segmentation in fetal MRI using convolutional neural networks. Magnetic resonance imaging, 64, 77-89.

[22] Anquez, J., Angelini, E. D., & Bloch, I. (2009, June). Automatic segmentation of head structures on fetal MRI. In 2009 IEEE International Symposium on Biomedical Imaging: From Nano to Macro (pp. 109-112). IEEE.

[23] Huisman, T. A., & Kellenberger, C. J. (2008). MR imaging characteristics of the normal fetal gastrointestinal tract and abdomen. European journal of radiology, 65(1), 170-181.

[24] Justice, R. K., Stokely, E. M., Strobel, J. S., Ideker, R. E., & Smith, W. M. (1997, April).

Medical image segmentation using 3D seeded region growing. In Medical Imaging 1997: Image Processing (Vol. 3034, pp. 900-910). International Society for Optics and Photonics.

[25] Tourbier, S., Velasco-Annis, C., Taimouri, V., Hagmann, P., Meuli, R., Warfield, S. K., ... &

Gholipour, A. (2017). Automated template-based brain localization and extraction for fetal brain MRI reconstruction. NeuroImage, 155, 460-472.

[26] Somasundaram, K., & Kalavathi, P. (2014). Brain segmentation in magnetic resonance human head scans using a multi-seeded region growing. The Imaging Science Journal, 62(5), 273-284.

[27] Vijayalakshmi, S., & Durgadevi, P. S. N. P. (2016). A Simple Approach For Relatively Automated Hippocampus Segmentation From Sagittal View of Brain.

[28] Somasundaram, K., Gayathri, S. P., Rajeswaran, R., & Manjiri, D. (2018). Fetal brain extraction from magnetic resonance image (MRI) of human fetus. The Imaging Science Journal, 66(3), 133-138.




Related subjects :