Int. J. Communications, Network and System Sciences, 2010, 3, 801-804
doi:10.4236/ijcns.2010.310107 Published Online October 2010 (http://www.SciRP.org/journal/ijcns)
A Secure Transfer of Identification Information in
Medical Images by Steganocryptography
Shuhong Jiao1, Robert Goutte2
1Information and Telecom Department, Harbin Engineering University, Harbin, China
2Lab. CREATIS, UMR CNRS 5520, INSERM U 630, INSA, Université of Lyon, Bâtiment Leonard de Vinci , 21 avenue
Jean Capelle 69621 VILLEURBANNE Cedex, France
Received July 26, 2010; revised August 23, 2010; accepted September 25, 2010
Abstract
The fast growth of the exchange traffic in medical imagery on the Internet justifies the creation of adapted
tools guaranteeing the quality and the confidentiality of the information while respecting the legal and ethical
constraints, specific to this field. The joint usage of steganography and cryptography brings an efficient solu-
tion, whose implementation in medical routine is realistic, thanks to the current progress in data processing
(broad band Internet access and grid computing).
Keywords: Medical Image, Identification, Steganography, Cryptography
1. Introduction
The current needs in medical imaging security comes
mainly from the development of the traffic on Internet
(tele-expertise, telemedicine) and to establishment of
medical personal file [1]. Among all possibilities it is
interesting to work on the messages concealment in the
image itself, then regarded as a medium coverage.
2. Objective
Insert in a medical image 2D black and white of any
modality, a hidden message with all information identi-
fication (the radiologist and patient), historical of record,
parameters of examination (nature, location, diagnosis
and comments of the radiologist). Ideal characteristics
sought:
1) Access, in reception, at the original image, without
alteration or loss of information
2) The method used for the encryption must resist to
attacks.
3) The hidden message must be, in reception, readable
by the holder of the key.
This group of ideal conditions, with contradictory’s
imperatives, will be practically never satisfied. However;
the proposed methods must be realized with objectives
neighboring of these limits.
3. Features Specific Constraints in Medical
Imagery
1) Need to use standards [2]
2) Respect of legislation
3) Rapidity and simplicity of implantation
4) Compatibility with JPEG compression
It is important to note that, in medical imagery domain,
the compression, to be truly operational, shall keep use-
ful information for th e diagnosis and should relate essen-
tially optimization of acquisition parameters, noise re-
duction and elimination of temporal redundancies.
4. Steganography
The steganography (Greek steganos: Covered and Gra-
phein: Write) is the art hide a message in a medium cov-
erage (medical image for example), so no one can dis-
tinguish the medium (original image), after the inclusion
in the hidden message [3].
The hidden message can be a plain text or his en-
crypted version. In this latter case (which is interesting
here), we use the term steganocryptography. For this, we
use a prior encryption of the hidden message, before the
introduction in the original image, considered here as a
medium coverage. We propose to use a symmetric en-

S. H. JIAO ET AL.
802
cryption algorithm, known as international standard. En
agreement with the work of W. Puech and M. Rodrigues
[4], we choose the algorithm AES. This symmetric ci-
pher uses blocs of data swapped of 128 bits and key sizes
of 128, 192 or 256 bits.
4.1. Example of AES Encryption and Decryption
[2]
Password: Creatisuniversitedelyon
Plaintext: Secure transfer in medical imagery
Encrypt it:
z4USOP2/O1sZdydqd4hSddOQUfRQabfKUCAoJP2drF
6entGV
Decrypt it: Secure transfer in medical imagery
4.2. Conversion of this Encrypted Message in New
Digital Message, ASCII 8 Bits by Character
[5]
z4U50P2/O1sZdydqd4hSdd0QU………
01110100011010001010101001101010100111101010000
00110010001011110100111100…..
5. Insertion by Steganography of Digital
Data in Original Digital Image
These methods require five successive steps:
The first step is to divide the image into 8 × 8 square
blocks (one byte for one grey level). In the second step
we compute the different DCT coefficients (Discrete
Cosine Transform [6] of these different blocks.
We select, in the third step, two spectral coefficients:
(amn) and (akl) in the block i. Their location, in this block
requires 2 × (2 × 3) bits. These 12 bits, expressed with
two ASCII characters (6 bits per character) are the be-
ginning of the share d hi d de n key .
In the 4th step we use the following rule: If bi = 1 and
(akl)  (amn) or if bi = 0 and (akl )  (amn) nothing is
changed .If these conditions are not carried out we ex-
change the values of (akl ) and (amn).
The amplitudes of the change of the spectral coeffi-
cients can be adjusted depending on the level of noise
and the rounding’ s error. The frequ ency’s position of the
two coefficients is important. If it is located in BF, the
method is robust, but risk of be visually detectable.
Instead, if it is located in HF, the original image will
be virtually unchanged, but the method will be more sen-
sitive to photometric fluctuations.
5.1. Variant
If necessary, for obtain another form of resistance to at-
tacks, we can take different coordinates for the coeffi-
cients (amn) and (akl). In this case, it is possible, for ex-
ample, with one bit, of displace the sequence for the
block i, for obtain the sequence of the block i + 1.
If coordinates (akl)i = 010 and 011 ,and if coordinates
(bmn)i = 001 and 100 we obtain the key chain for the
block i: 010011001100 and 001001100110 for the block
i + 1.
The detection of these coordinates is more difficult,
but it is not possible to choice a single optimal domain
for these coefficients, in the spectral plan.
We have the ability to hide 1 bit per squ are block or, for
an original image of size 1024 × 1024, to hide 16384
bits .With ASCII code ,extension UNICODE (with 8bits per
characters),we obtain the possibility to hide 2048 characters
in this image.
The 5th step is the extraction of the hidden message.
The meth od is simil ar to that th e insertion : at the rec ep-
tion we compare the values of the two selected coeffi-
cients The previous rule allows us to know if the bit
concerned is 1 or 0.
Remark: The marking brought a very slight loss of in-
formation, since the image in the reception is not exactly
identical to original, but the quantity of bits transferred
is the same.
It is important to note this favorable factor: Any loca-
tion in the spectral domain involve an displaying in
the image plane, which facilitate the invisibility of the
insertion
6. Radiographic Application
The original image (Figure 1) is a pulmonary radiogra-
phy, obtained in to modens itometry (X ray scanner).
We isolate on this image one block 8 × 8.
Figure 1. Pulmonary scannography

S. H. JIAO ET AL.
803
142 120 100 87 82 78 79 81
131 113 98 87 79 83 82 82
119 107 97 90 84 83 82 80
119 112 106 100 95 85 83 80
134 127 118 107 100 91 87 82
150 140 126 111 100 96 91 85
156 144 129 114 103 96 92 86
145 132 119 108 100 91 88 83
(Original block image, before transform)
After Discrete Cosine Transform DCT, we obtain this
block, in the spectral domain.
3.2240 0.5578 0.1552 0.0572 0.0132 0.0177 0.00200.0074
0.2305 –0.0818 0.0513 0.0124 0.0143 0.0023 0.0074–0.0042
0.0148 0.0510 0.0676 0.0211 0.0094 0.0037 0.0036–0.0020
0.0986 0.0704 0.0268 0.0042
–0.0060 0.0037 0.00140.0002
–0.0270 –0.0070 0.0226 –0.0046 0.0152 –0.0070 –0.0071 0.0109
0.0156 0.0113
–0.0051 –0.0028 0.0062 –0.0026 –0.0046 0.0044
–0.0194 –0.0052 0.0026 –0.0027 0.0009 –0.0005 –0.0010–0.0001
0.0048 0.0046 0.0007 0.0010 0.0002
–0.0011 –0.0013 0.0017
With k = 3, l = 2 and m = 3, n = 3, akl = 0.0510, and
amn = 0.0676
Here: akl = 0.0510 < amn = 0.0676
If I want to introduce a hidden bit b = 0, nothing is
changed.
If I want to introduce a hidden bit b = 1, we invert the
values of akl and amn
akl = 0.0676 > amn = 0.0510
In this case (b = 1) we obtain after inverse transforma-
tion DCT-1
142 120 101 88 83 78 78 79
131 113 98 87 79 83. 82 81
119 107 97 90 84 83 82 81
119 112 105 99 94 85 84 82
134 127 117 106 99 91 88 84
150 140 126 111 100 96 91 86
156 144 129 114 103 96 92 85
145 132 120 109 101 91 87 81
(bloc image after crypt)
On each pixel modified, the positive or negative error
is approximately one grey level. Only 24 pixels dis-
persed on 64 are modified and the mean of grey levels
is unchanged. The image is good and the message is not
visible. Practically, in a medical image with 256 gray
levels, a deviation of one pixel has no physical signifi-
cance and presents no interest for visual observation or
subsequent numerical processing. It results mainly from
the presence of noise and artifacts of rounding obtained
when quantifying.
To avoid the consequences of too large distance be-
tween these 2 factors (akl and amn) which may introduce a
important disturbance of the spectrum) and also the op-
posed consequence of a gap too low (rendering the method
unstable in presence of noise) we use, if necessary the
following rule:
If the gap amn – akl is 2e, and if 0.0010 < e < 0.0120,
nothing is modify;
If e < 0.0010 we take akl’ = 0.5(akl + amn) – 0.0010
and amn’ = 0.5(akl + amn)+ 0.0010
If e > 0.0020 we take akl’ = 0.5(akl + amn) – 0.0120
and amn’ = 0.5(akl + amn) + 0.0120
0.0010 and 0.0120 are too adjustable parameters, in
function of the leve l of noise in the image and the co nfi-
dential degree wished.
6.1. Example of Hidden Message
Place of examination: Cardiologic Hospital, HEH Lyon,
Service of R a diology.
Date: 25/15/2009
Instrumentation: Tomodensitometer,
Pulmonary Scanner
Radiologist: Dr. Jean Martin
Identification : XXXXXXX
Patient: Michel Dupont
Identification : YYYYYYY Age:45 years
Conditions of observation: Axial
Incidence transverse
Commentaries: This patient presents a small paren-
chymatous nodule of the higher segment of the lobe
lower right. Presence also of a discrete bronchial dila-
tion in the average lobe.
This hidden message (in italic) possesses 347 charac-
ters, with spaces. In ASCII we obtain 1.421.312 bits, it is
necessary to have a dimension for the original image
equal or upper of 1.5 megabits. This dimension is usual
in medical imagery.
Thus, in our example, the proposed coding is invisible
and involves no loss of us eful inf or matio n.

S. H. JIAO ET AL.
804
This approach has been submitted to a panel of radi-
ologists from hospital, specialists from different imaging
modalities and their comments and proposed additions
have been included in this final implementation.
7. Generalization
The method proposed is well suited to the JPEG com-
pressed images [7] because, in this case, the compression
algorithm use also the Discrete Cosine Transform (DCT).
It is possible to extend this method to color images,
which can be considered as a set of three images black
and white (RGB). In three dimensional imaging, we can
consider 8 × 8 × 8 cube and use the 3D block DCT algo-
rithm.
8. Conclusion
The steganocryptography can transmit, with invisibility
and robustness, the information accompanying a medical
digital radiography. The constraints of legal requirements,
safety and confidentiality are fully satisfied.
9. References
[1] Liliane DUSSERRE, Rapport du Conseil National de
l’Ordre des médecins, France, 2002.
[2] Federal Information Processing Standards, Public. 197,
announcing the Advanced Encryption Standards (AES),
USA, 2001.
[3] Christian REY, Jean Luc DUGELAY, Panorama des
méthodes de tatouage, Traitement du Signal, Vol. 18,
spécial No., 2001.
[4] W. Puech and J. M. Rodrigues, “Crypto-Compression of
Medical Images by Selective Encryption of DCT,” 13th
European Signal Processing Conference, EUSIPCO’05
Antalya, Turkey, 2005.
[5] Wikipedia, Encyclopédie libre, Norme ASCII.
[6] Y. Wang and P. Moulin, “Steganalysis of Block-DCT,
Image Steganography,” Proceedings of IEEE Workshop
on statistical Signal Processing, St Louis, 2003, pp. 339-
342.
[7] H.-W. Tseng and C.-C. Chang, “Steganography Using
JPEG Compressed Image,” Fourth International Confer-
ence on Computer and Information Technology (CIT 04),
2004, pp. 12-17.