INTRODUCTION

Contrast is introduced into the blood stream for enhancement of brain tumors and various lesions depending upon the degradation of the blood-brain barrier. Iodine has  Atomic No.(Z-53) CM is radio-opaque substances that can be administered intravenously or orally into the body, in order to enhance the radiographical visualization of internal structures (1). Iodinated CM show higher attenuation levels at lower X-ray tube voltage owing to higher photoelectric effect and lesser Compton scattering (2,3,4). The mean photon energy of the tube potential is adjusted closer to the k edge of iodine. The influence of the photoelectric interactions in the present of iodinated CM is greater compared with Compton scattering effects because of the 33-keV k-edge of iodine (5). As a consequence, it leads to an increase in the linear attenuation coefficient of iodine much more dramatically than water and this causes excellent  contrast enhancement (Figure2a and b) (6,7).

Figure 1

Pre and post contrast CT scan of tissue showing differences in their CT Value

Figures 2ab

A. Plain CT scan of Meningioma B. Post contrast CT scan.

Iodinated medium may also be either ionic or non-ionic. The ionic type tends to create a high osmolality  in blood and may cause a contrast media reaction in some individuals, which may be life-threatening for the sensitive patients. The non-ionic form decreases this risk, but is much more expensive. The non-ionic contrast media is much more widely used today (9).  Iodinated contrast causes contrast mediated nephropathy (CMN) and systemic fibrosis and irreversible complications(10,11,12,13). Kidney function of the patient receiving a dose should be considered before the exam (9,14,15) The estimated glomerular filtration rate (eGFR) should be no lower than 30 mL/min in patients receiving iodinated contrast(16,17). As both plain and contrast enhanced studies are done there is chance of excessive radiation. CT has a typical effective dose (of whole body) of 2mSv(18,19),ie equivalent to 200 chest X ray or 243 days natural background  radiation  compared to the  typical absorbed dose of 56mGy of the brain.  

The use of CM during CT scans was found to significantly increase the radiation dose present in several organs, a new study indicated that both the iodine quantity and the number and timing of contrast-enhanced CT scans should be optimized when setting protocols(19).

Radiation hazard may lead to malignant disease though malignancy is a stochastic effect of radiation, meaning that the probability of occurrence increases with effective radiation dose, but the severity of the cancer may be  independent of dose. As per American college of Radiology CT head for various lesion without and with contrast is criteria 4 (19,20).

2) Drug reaction or adverse effect - Minor allergic reaction to life threatening drug reaction can happen and one death out of 200,000 contrast injection  can occur(20).

X Ray-beam is attenuated by tissue in CT Scan is mainly by Compton Scattering. The definition for Linear Attenuation Coefficient (given the symbol μ) is the percent reduction per unit thickness of absorber material(tissue) .  or  where is the intensity of emergent beam, is the intensity of initial beam and is the thickness of tissue.

CT number  explains the tissue Character.

CT NUMBER =              HU=Hounsfield unit

water m                                             

Assessment of contrast enhancement on normal and various brain lesions was the essential element of our study. Statistical analysis of the pre- and post- contrast brain lesions, the intensity of CTnumber( signal) in gray shades and the extent of contrast enhancement in tissue depend on the iodine content within each pixel (20). We measured the CT number in gray and white matters and in brain tumors before and after administration of a contrast . Relative electron density is the ratio of electron density(N) of the tissue to the N of water. Thus water with a physical density of 1 and N of 3.340X10¹⁵ /cm³ has a relative electron density of 1. For a CT image of a mass or lesion in the brain containing unknown materials, we calculate the relative electron density for each pixel using “relative electron density-CT Number relations curve” (Figure3) provided by Watanabe and Matsufuji et al (21,22,23,24).  N of the tissue in question can be obtained by multiplying relative electron density by 3.340X10²³ cm³ (25)

Our purpose was to assess whether there was any statistical relationship in the magnitude of changes in gray scale value and electron density of brain lesions or  tumors before and after the contrast material was administered, and whether enhancement-like effects of the lesions could be reproduced mathematically with the help of neural networks and without the IV  contrast injection(26).

CT Number, electron density and signal intensity, as represented by the corresponding gray shade values (out of 256 shades) of various brain lesions before and after IV contrast were tabulated. Using independent component analysis (ICA), a statistical technique, the complex dataset was decomposed into independent sub-parts (26).

With the help of K-mean clustering, the data were classified. A mapping function was generated that corresponded between these independent components and the cross-sectional data by using the neural network after training it with sample datasets (26,27,28,29). The training samples for the network were selected using K–mean clustering.  Mapping of the lesion was overlaid on the plain or non contrast image as contrast enhancement.

MATERIALS AND METHODS

REQUIREMENTS FOR SIMULATION TECHNIQUES

A. Pre Contrast CT value of tissues

B. Pre contrast signal shade(intensity) in the image pixel

C. Pre Contrast electron density of tissue

D. Post contrast CT value of tissues.

E. Post Contrast signal shade in the image pixel .

F. Post Contrast electron density of tissue

G. Neural Tool to live predict the shades of unknown tissue 

CT imaging data were collected from normal brain as well as from the pathological cases. After obtaining proper institutional ethics, 81 normal control cases and 21 pathological cases were studied. Pathological cases included low grade glioma, glioblastoma,, solid brain abscess, tumefactive multiple sclerosis, lymphoma, metastasis from breast and lung cancer and neuro-cysticercosis (Table1).

Table 1

The CT Numbers for normal brain tissue, as well as for various brain tumors and lesions, were determined in a 16 slice multi Detector Toshiba (Japan), Alexion CT scanner. These CT values were compared with the gray value of the CT images, using a matrix size of 512 × 512 and a slice thickness of 5 mm. Electron density or N of normal and abnormal gray white matter were determined using standard mathematical expressions directly from the CT Number of the image pixel of CT image.

The linear attenuation coefficient (μ) has a linear relationship with the density of tissue. Relative electron density is the ratio of N of the tissue to the N of water. Thus water with a physical density of 1 and N of 3.340X10¹⁵ /m³ has a relative electron density of 1. For a CT image of a mass or lesion in the brain containing unknown materials, we calculate the relative electron density for each pixel using “relative electron density-CT Number relations curve” (figure3) provided by Matsufuji et al and Watanabe . N of the tissue in question can be obtained by multiplying relative electron density by 3.340X10²³ cm^-3. (Table2).

Figure 3

CT Number to relative electron density conversion curve. Relative electron density is electron density (N) relative to N of water with density of 1.Multiplying relative electron density with 3.340X1023 /cm3 of N of the tissue is derived.

Table 2

Tissue CT No electron density in plain and post contrast state

After that each patient received via infusion: 120-250 ml (Omni 240mg/ml) by a power injector (Magmedix) at a high-flow rates up to 5 ml/second and was imaged by multi detector sixteen slice  Alexion CT scanner( Toshiba) applying 120KV, 200 mAS and 5mm slice thickness.

Regions of interest (ROI) representing enhancing and/or non-enhancing parts of brain lesion or tumors were selected by viewing the post contrast plain images. ROI containing white and gray matter areas were identified in the   contra lateral hemisphere. CT values of the various part and pixel of the brain lesion and tumor were directly measured from ROI in a special DICOM editor (Sante’ DICOM Editor)  (Figure4 a and b)

Figure 4a

Plain and b. post contrast CT scan image in a DICOM editor. ROI showing pixel values.

SEGMENTATION & K MEAN CLUSTERING:

 The pre and post contrast images were then decomposed as mosaic pattern (Figure 5) so that various pixels with different gray shades can be appreciated. Gray shade value in Gray shade scale out of 256 shades (2⁸ bit) in 16 different slabs were determined and tabulated. Relationship of shades and CT value is shown in figure 6. By applying K-mean clustering the gray shade values( out of 256) of various parts were determined and tabulated. K- mean clustering classified and data based on attribute/features. Each attribute represented the  CT value of tissue before and after contrast and gray shade value as the signal. The grouping is done by minimizing the sum of square of distances between data and the corresponding cluster centroid. Then pre- and pos contrast values of the tissues were transferred into the excel sheet of the neural tool(30,31)

Figure 5

Mosaic pattern of pre and post contrast image showing pixels of gray shades, ROI mentioning pixel values.

Neural Network :

Neural Network Design and Steps

We followed the standard steps for designing neural networks to predict the signal and gray shade values of unknown tissues in four application areas: function fitting, pattern recognition, clustering, and time series analysis (26). The work flow for any of these problems had seven primary steps:

 i) To collect data from Plain CT images and k mean clustering.

ii) To create the network

iii)To configure the network

iv)To initialize the weights and biases

v)To train the network

vi)To test and Validate the network

Table 3

CT number(values) plain(CT No.1) post contrast (CT No.2) of various Tissues in excel data sheet of Neural Tool

Figure 6

Range of CT number( value) and corresponding gray shade.

To use the network for prediction of value of unknown tissue.  A data set manager is prepared in Microsoft Excel: neural tool that contain 5 variables, 32 cells per variable containing known data  neural network program. Neural tool 6.1.2   of the Palisade Inc was used which initiated the brain-like functions in order to learn the structure of the data provided(26).

Figure 7

Dataset manager in Neural Tool

Out of 5 variables:-

A. One independent category variable (of tissue),

B. Three independent numeric category variables such as CT values  i) Pre-contrast ii) post-contrast Tissue iii) gray shade signal value of pre- and post-contrast tissue (out of 256 shades)

C. One dependent numeric category variable, which ultimately produced live predict variables of post contrast values.

After understanding the data, the program worked through the following steps: analysis of the data, training the network on the data provided  and making live  intelligent predictions of the signal (gray values of the pixel out of 256 shades) from the new input of the plain DICOM  CT scan image of the unknown image(target tissue) .  

Figure 8

Training and Testing of the data by Neural Tool

Figure 9

Live prediction of unknown tissue/tumor by Neural Tool.

Contrast Enhancement Simulation

A radial basis function network was then used to generate a contrast enhanced map overlay on the background of plain DICOM image. It produced correspondences between independent components of gray shades of the pre-contrast plain image and live prediction of the dependent variable of gray shades (out of 256) of tumor/lesion of the plain DICOM image, pixel by pixel. To make the procedure simple a palette was prepared from the pre contrast gray shade value with corresponding post contrast value. By running the color scheme program of the DICOM editor (Santé Soft) of contrast enhancement in the DICOM editor a contrast like effect of the image of the brain lesion or tumor occurs.

Results

The mean CT values in tumors was 44 to 54 HU before administration of contrast, and 44 to 58 HU after contrast. Meningioma shows increased CT value of 60-62 HU which after contrast becomes 68 to 70 HU . CT value of white matter is noted between 16 to 22 HU which after contrast increased up to 18 to 24 HU. In plain study gray matter has value of 26 to 37 HU where as post contrast image shows value between 29 to 30 HU.

Figure 10

Relationship between Pre and Post contrast CT values

Discussion

Analysis of CT values.

Detectable tumor enhancement was noticed in all 21 patients. After the contrast injection, CT number increased as mentioned earlier in the result section. Slight increase in the gray and white matters were noticed. No enhancement of the vasogenic edema and perilesional edema were noted. Analyzing the percentage of  increase in CT number after contrast enhancement, it is evident that it is at a minimum in white matter (0.15 to 12.84%, mean-3.3%) and maximum (79%, mean) in case of tumors.  From the relationship shown in Figure10 it is noted that,

Post-contrast CT value = 1.061 x +0.050 X Pre-contrast CT value    (Equation 1)

Figure 11

A and B Relationship of the post-contrast CT value and signals expressed in gray shades

Figure 11 depicts the relationship of the CT value  and post-contrast signals expressed in gray scales:

CT value = 0.379XGray shade value - 2.853 (Equation 2).

Neural Network

We used the General Regression Neural Networks (GRNN) numeric predictor configuration.

Architecture of a PNN/GRNN Network:

A method was proposed to formulate the weighted-neighbor method described above in the form of a neural network (24-27). It was called a “Probabilistic Neural Network.” Here is a layer scheme of a PNN/GRNN network (28,29):

LAYERS:  INPUT NODES--àHIDDEN NODES—>CLASS NODES—>DECISION NODES(PREDICTION)

Figure 12

Layers of neural network

All PNN/GRNN networks have four layers: input, hidden, pattern/summation and decision/prediction.

Input layer — Pre- and post-contrast T1 values of gray and white matters, various tumors and brain lesions.

The input neurons/nodes (or processing before the input layer) standardize the range of the values by subtracting the median and dividing by the interquartile range used to summarize the extent of the spread of the data (variability between 75 th and 25 th percentiles).

 The input neurons then feed the values to each of the neurons in the hidden layer (30). In table 3, a summary of the neural network is provided.

Table 4

Summary of The Neural Net

Table 4 depicts how the dataset manager trains the inputs and tests to predict. Most of the predictions were found to be good. Residual values (between original and predicted value) show a range from -0.15 to 4.25.  There are 5 variables and 24 data cells per variable.

Hidden layer — There were 19 cases are used during training and  trials. During testing, the tool tried 5 predictions, out of which a 0 % were found bad after analyzing the residual values (actual values minus predicted values).

The trained Neural Net’s prediction (auto-testing) of unknown post-contrast CT values of pathological tissues when pre-contrast CT values and signal (gray) shades were provided are in Figure9. Below is the table for R-Square (Training), Root mean Sq error (training), Root mean Sq (Testing)  along with linear function.

Table 5

R-Square (Training), Root mean Sq error (training), Root mean Sq (Testing) And linear function of tissues

Figure 13 A,B and C depict the relationship of Predicted VS Actual(testing),Residual VS actual testing and predicted VS actual training.

Figure 13a

A and B discern the relationship of Predicted versus Actual, Residual versus Actual (Training) and Predicted versus Actual testing.

Figure 13b

A and B discern the relationship of Predicted versus Actual, Residual versus Actual (Training) and Predicted versus Actual testing.

Figure 13c

Predicted Vs Actual training

Simulation of Contrast enhancement

Figures 14  and 16 A, B, and C show  a meningioma in the plain , contrasted  image and brain mapping of contrast-like effect. Contrast simulation and original iodinated contrast enhancement have similarities in shape and pattern.  15A and B are the images of contrast like effect simulated in a case of glioblastoma.

Figures 14abc

show the non-contrast Meningioma, contrasted image(slightly higher position) and brain mapping of the contrast-like effect overlay on the plain image

Figures 15ab

Simulated enhancement of Glioblastoma in the right parieto-occipital lobe.

Figure 17 depicts Glioma  and Figure18 shows pineoloma in plain, post contrast and contrast simulation study.  In some areas due to high signal noise ratio quantum mottles are seen.

Figure 16

Plain, post contrast and contrast enhancement simulation of meningioma.

Figure 17

Plain and simulated contrast enhancement of glioma in the right frontal lobe

Figure 18

Pineoloma – A. Plain B. Post Contrast, C. simulation effect

DICOM FORMAT VS JPEG FORMAT

CT scan  images are displayed in Digital Imaging and Communications in Medicine (DICOM) format for simulation effect compared to exported images in JPEG  (lossy 8 bits/pixel) compression) format.  The DICOM image is also a 8-bit gray scale image (signed/unsigned). This means each pixel is represented by 8-bit or 2⁸ = 256 shades/pixel of gray level combinations available, so a smooth and regular outline is depicted in the simulation effect unlike the JPEG images.

Conclusion

An attempt was successfully made to generate an enhancement like effect in various brain lesions statistically, utilizing a neural network and without the IV contrast.