Enhancing Laryngeal Spinocellular Carcinoma Image Security with DCT

Abstract

This work proposes a novel secret sharing scheme to enhance the security of Laryngeal Spinocellular Carcinoma or Laryngeal Squamous Cell Carcinoma (LSCC) images using the Discrete Cosine Transformation (DCT) as a cryptographic tool. The DCT-based secret sharing method divides LSCC images into shares, each containing DCT coefficients that represent the image’s frequency components. The original image can only be reconstructed when a predefined number of shares are combined, ensuring confidentiality and preventing unauthorized access. The proposed scheme demonstrates robustness against noise and data loss during transmission, preserving image quality and data integrity. The performance of the proposed scheme concerning the quality of the recovered image and the strength of security preservation is demonstrated through PSNR improvement analysis, correlation analysis, and histogram analysis. The efficiency of DCT-based secret sharing enables application in medical settings, facilitating accurate diagnosis and treatment planning for LSCC patients while safeguarding patient privacy.

Introduction

Laryngeal Spinocellular Carcinoma (LSCC) is a severe and life-threatening form of cancer affecting the larynx, necessitating accurate diagnosis and prompt treatment through medical imaging. In the context of secure data sharing and privacy protection, this study proposes a novel secret sharing scheme using the Discrete Cosine Transformation (DCT) as a cryptographic tool to enhance the security of LSCC images.

The Discrete Cosine Transformation (DCT) offers several strengths in image processing and steganography, making it particularly attractive for secure data sharing applications. One of the key advantages of DCT lies in its energy compaction properties, which efficiently represent image content by concentrating most of the signal’s energy into a few low-frequency coefficients [1]. This attribute is highly beneficial in data compression, enabling efficient representation of images with minimal data redundancy, and reducing the data size, which is crucial in the context of secure image sharing and storage [2].

Furthermore, DCT operates on real-valued data and produces real-valued coefficients, simplifying computations and reducing computational complexity in image processing tasks [3, 4]. This real-to-real transformation is advantageous for steganography, as it ensures that modifications to the image data are less likely to introduce perceptible artifacts, enhancing the imperceptibility of hidden data [5, 6]. Additionally, the block-based processing nature of DCT allows for parallelization and efficient hardware implementations, enabling faster processing times and real-time capabilities, a crucial aspect in secure data sharing applications, where timely and efficient encryption and decryption are paramount [7, 8, 9].

In this study, we explore the potential of the DCT-based secret sharing scheme to securely divide LSCC images into shares and distribute them among authorized parties, safeguarding patient confidentiality and facilitating secure collaborations among healthcare professionals and researchers. By leveraging the strengths of DCT in image processing and steganography, this research aims to enhance data privacy and advance medical research and personalized treatment strategies for Laryngeal Spinocellular Carcinoma. The flow of this study is organized as follows: Sect. 2 delves into the related works, while Sect. 3 comprehensively describes the proposed methodology. In Sect. 4, we present the experimental findings and results. Lastly, Sect. 5 offers a concise and insightful conclusion.

Related Works

Secure data sharing and privacy protection have been critical concerns in the field of medical imaging, prompting researchers to explore various cryptographic or secret sharing techniques for safeguarding sensitive medical image datasets [10]. Many secret sharing schemes result in degraded reconstructed image quality, discouraging their use in securing LSCC images. The original LSCC images already have poor quality due to imaging limitations [11]. Preserving diagnostic features is crucial, making it essential to develop secure schemes that minimize information loss during sharing and recovery. Balancing security and image quality is key for effective and reliable protection of sensitive medical imaging data [12].

In the context of image processing and steganography, the Discrete Cosine Transformation (DCT) has emerged as a prominent tool offering several strengths [13, 14]. Recently, Fan et al. [15] investigated the application of DCT in medical image watermarking, demonstrating its robustness in embedding imperceptible watermarks for image authentication and tamper detection. The real-valued nature of DCT coefficients ensures minimal distortion to the original medical image, making it suitable for medical data authentication in secure environments [16, 17]. Researchers have also explored the potential of DCT in image compression, especially in the JPEG (Joint Photographic Experts Group) image compression standard [18, 19]. Ravi Kumar et al. proposed an optimized JPEG compression technique using DCT for efficient storage and transmission of medical images while maintaining diagnostic quality. The energy compaction properties of DCT enable high compression ratios, making it an ideal choice for reducing data size and facilitating secure image sharing while ensuring minimal loss of crucial medical information [20].

In the domain of secure data sharing, A. Tiwari et al. developed a novel secret sharing scheme based on DCT for secure transmission of medical images in telemedicine applications. The researchers showcased that block-based processing with DCT-based techniques requires real-time capabilities for practical applications in telemedicine. Furthermore, AJ Fofanah et al. explored the application of DCT in steganography to embed patient data and annotations into medical images while preserving image quality and ensuring secure communication [14]. The real-valued outputs of DCT facilitated imperceptible data hiding, making it a promising technique for privacy-preserving medical image sharing and collaborative research efforts among healthcare institutions [21].

In summary, the literature survey highlights the growing interest in leveraging the strengths of Discrete Cosine Transformation (DCT) in image processing and steganography for secure data sharing and privacy protection in medical imaging. The energy compaction properties, real-valued outputs, and block-based processing nature of DCT offer significant advantages in medical image compression, authentication, and data hiding. Researchers have explored DCT-based techniques for secure transmission, telemedicine applications, and collaborative research efforts, emphasizing the need for optimization to achieve real-time capabilities and practical implementation in healthcare settings. The proposed DCT-based secret sharing scheme in this study aims to contribute to the existing body of research by enhancing data privacy and fostering secure collaborations in the context of Laryngeal Spinocellular Carcinoma (LSCC) images.

Proposed Methodology

Figure 1 depicts the proposed methodology pipeline. In this proposed methodology, we aim to securely share and recover Laryngeal spinocellular carcinoma (LSCC) images among multiple participants while preserving the image quality and confidentiality. The process involves two stages: Share Creation and Reconstruction.

Fig. 1.

Block diagram of the proposed scheme using DCT

In the Share Creation stage, we begin by converting the original LSCC image into halftone representation, which simplifies the sharing process by converting continuous-tone pixels into binary (black and white) pixels. From this halftone image, we retain the low-resolution version, as it will be embedded into the shares later. We then divide the halftone image into l blocks, where l represents the size of the secret image divided by the number of participants (n) in the sharing scheme. To generate the shares, we create $(n+1)$ basis matrices denoted as $M^0$, $M^1$,..., $M^n$, as illustrated in Table 1 for the case where $n=4$.

Table 1.

Basis matrices for $n=4$ Shares

$M^0=\left[\begin{array}{cccc}0& 0& 0& 0\\ 1& 1& 1& 1\end{array}\right]$	$M^1=\left[\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 1& 1\end{array}\right]$
$M^2=\left[\begin{array}{cccc}0& 1& 0& 0\\ 1& 0& 1& 1\end{array}\right]$	$M^3=\left[\begin{array}{cccc}0& 0& 1& 0\\ 1& 1& 0& 1\end{array}\right]$
$M^4=\left[\begin{array}{cccc}0& 0& 0& 1\\ 1& 1& 1& 0\end{array}\right]$

In the secret sharing scheme, every basis matrix corresponds to a distinct share, with the exception of $M^0$, which is specifically used for the White pixels of the secret image across all shares. Importantly, we add an element of randomness by selecting a row randomly from each basis matrix. This randomness increases the security and robustness of the sharing process.

Using these basis matrices and block-wise pixel values, we generate n shares, each of equal size to the secret image. The shares presented in Fig. 2 are designed in a manner that ensures they do not individually reveal any secret information. Their construction is such that only by combining a sufficient number of shares can the original secret image be reconstructed and its content revealed.

Fig. 2.

Representative four shares

Next, we transform the n shares into the $8\times 8$ block Discrete Cosine Transform (DCT) domain, which allows for efficient processing and analysis of the image data [22]. The high-resolution frequency components of the $8\times 8$ DCT blocks are quantized to reduce the data size using Table 2.

Table 2.

Proposed Quantization Table

8	10	12	14	16	18	20	22
10	12	14	16	18	20	22	24
12	14	16	18	20	22	24	26
14	16	18	20	22	24	26	28
16	18	20	22	24	26	28	30
18	20	22	24	26	28	30	32
20	22	24	26	28	30	32	34
22	24	26	28	30	32	34	36

Subsequently, the low-resolution halftone image is embedded into the middle-frequency components of these blocks using a secret key. Embedding in the middle frequency components ensures a better quality of the recovered image during the reconstruction stage and protects the secret information from being perceptible to the Human Visual System (HVS) [23].

Moving to the Reconstruction stage, we begin by extracting the embedded low-resolution secret image pixels from the middle-frequency components of the $8\times 8$ DCT blocks using the secret key. The extracted pixels are then upscaled using interpolation to obtain the high-resolution version of the secret image. Next, we perform the inverse DCT on the shares to convert them back to the spatial domain. To recover the secret image, we perform an XOR operation between all the shares along with the high-resolution image obtained in the previous step. This XOR operation results in the reconstruction of the original secret image. To improve the quality of the recovered image and remove any noise effects and blocking artifacts introduced during the sharing and reconstruction processes, we apply a Gaussian filter. This step ensures that the final recovered image closely resembles the original LSCC image, providing accurate diagnostic information while maintaining the confidentiality of sensitive medical data.

Experimental Findings and Results

The image dataset available at [24] was utilized to develop and evaluate the proposed model, employing Python and OpenCV for implementation. The dataset consists of 1320 patches extracted from narrow-band laryngoscopic images of 33 patients diagnosed with laryngeal spinocellular carcinoma after histopathological examination. These patches, measuring 100x100 pixels, represent both healthy and early-stage cancerous laryngeal tissues. The dataset is composed of four distinct tissue classes, each containing 330 patches. The tissue classes considered are He (healthy tissue), Hbv (tissue with hypertrophic vessels), Le (tissue with leukoplakia), and IPCL (tissue with intrapapillary capillary loops). The manual extraction of these patches ensures accurate representation and labeling of the different tissue types for training and evaluation purposes. Due to space limitations, Fig. 3 displays one representative image from each of the four tissue classes: He, Hbv, Le, and IPCL.

Fig. 3.

Representative laryngeal spinocellular carcinoma tissue images

The performance of the proposed model is evaluated for both image quality and security robustness through PSNR (Peak Signal-to-Noise Ratio) improvement analysis, correlation analysis, and histogram analysis, as outlined below.

• PSNR improvement analysis The graph in Fig. 4 depicts the PSNR values for the proposed method. It starts with a single share PSNR of 12.6 dB. As the number of stacks increases from 2 to 4, the PSNR improves significantly, reaching 15.7 dB for 2 stacks, 18.5 dB for 3 stacks, and 21.9 dB for 4 stacks. This improvement in PSNR demonstrates the enhanced image quality and robustness achieved by increasing the number of stacked shares, indicating better fidelity and less information loss during the reconstruction process. The significant PSNR improvements validate the effectiveness of the proposed model in securing the Laryngeal Spinocellular Carcinoma images and ensuring reliable data sharing while preserving high-quality image information.

• Correlation analysis The correlation plot for 4 shares is shown in Fig. 5, where each share’s correlation with the other shares is visualized using a heatmap. The higher the correlation value, the stronger the relationship between the shares. The heatmap provides a clear visual representation of the correlations, helping to assess the interdependencies among the shares in the secret sharing scheme. The proposed work demonstrates more diverse and reduced correlation values between different shares. The lower correlation values among different shares, indicating a higher level of security and reduced interdependencies in the proposed secret sharing scheme.

• Histogram analysis

Fig. 4.

Improvements in PSNR with increasing stacks

Fig. 5.

Correlation plot for 4 shares with further reduced correlation

In the histogram comparisons, we present four distinct plots, each illustrating a comparison between different image pairs. Figure 6 and Fig. 7 showcase a noticeable misalignment between the share image and the embedded image concerning the original image. This observation underscores the security robustness of the proposed method, as intentional distortion of the share and embedded images ensures the protection of the secret information. Figure 8 provides evidence of the similarity between the share image and the embedded image, confirming the covertness achieved by the proposed scheme. This covertness is essential as it involves embedding secret information in the share image, which is crucial for the high-quality recovery of the hidden content during the secret reconstruction stage. Furthermore, in Fig. 9, we observe a remarkable overlap of the reconstructed image curve with the original image curve, indicating the superior quality of the recovered image. This outcome highlights the effectiveness of the reconstruction process, which successfully restores the original image’s visual fidelity during the secret recovery phase.

Fig. 6.

Histogram comparison - original image vs. share image

Fig. 7.

Histogram comparison - original image vs. embedded share

Fig. 8.

Histogram comparison - share image vs. embedded share

Fig. 9.

Histogram comparison - original image vs. reconstructed image

Conclusion

Our work presents a novel secret sharing scheme for secure Laryngeal spinocellular carcinoma (LSCC) image transmission, preserving quality and confidentiality. The proposed methodology includes halftoning, DCT-based embedding, and secret sharing techniques, enhancing image quality significantly. Additionally, the proposed quantization table facilitates efficient data compression and enhances the effectiveness of the image sharing process. The intentional misalignment between share and embedded images reinforces security robustness. The covertness achieved through embedding secret information within the share image ensures confidentiality. Successful reconstruction restores the original image’s fidelity, validating our approach. The scheme proves valuable for telemedicine applications, ensuring privacy and integrity of sensitive medical data. Future research can explore optimization techniques and real-world implementation.

Author Contributions

The first author led the design and implementation of the novel progressive image secret sharing algorithm based on the Discrete Cosine Transform (DCT), focusing on robust image recovery and secure distribution. The second author performed extensive performance evaluations validating its effectiveness in achieving superior image recovery.

Funding

Not applicable.

Data Availability

Available

Code availability

Available

Declarations

Conflict of interest

No conflict of interest.