Disease Prediction by Tongue Classification using CNN

Abstract

The integration of traditional medicine and modern technology has opened new avenues for disease prediction and diagnosis. In this study, we explore the use of tongue classification as a non-invasive and cost-effective approach to predict and diagnose various diseases. Tongue images were collected from a diverse patient population and processed to extract relevant features. Machine learning algorithms were employed to classify these tongue images into disease categories, yielding promising results. The study\'s findings demonstrate the potential of tongue classification as an efficient diagnostic tool, with implications for early disease detection and personalized healthcare. This research offers insights into the fusion of traditional medical knowledge with cutting-edge technology, highlighting the possibilities of enhancing healthcare through innovative interdisciplinary approaches.

Introduction

I. INTRODUCTION

In the quest for more effective and non-invasive methods of disease prediction and diagnosis, the confluence of traditional medical practices and modern technology has gained significant attention. Traditional Chinese medicine, with its rich history and holistic approach to healthcare, offers an intriguing pathway towards innovative healthcare solutions. One such avenue is the analysis of the tongue, a fundamental aspect of traditional diagnostic methods, which holds great potential in the realm of modern medicine.

The tongue, as a diagnostic tool, has been used in traditional medicine systems across various cultures for centuries. Traditional Chinese medicine, in particular, places considerable importance on tongue characteristics and their correlation with an individual's overall health. Changes in tongue appearance, such as color, shape, and coating, are believed to reflect underlying physiological imbalances or disease conditions. This ancient knowledge, passed down through generations, offers valuable insights into the body's health status.

In recent years, the integration of machine learning and image analysis techniques has allowed for the systematic quantification of tongue characteristics. These advancements provide a bridge between traditional wisdom and modern medical science. By utilizing computational methods to analyze tongue images, it becomes possible to not only document and quantify these characteristics but also to apply them in the context of disease prediction and diagnosis.

This research paper delves into the novel field of "Disease Prediction Using Tongue Classification." Our objective is to explore the potential of this integrated approach, where traditional tongue analysis is augmented by modern technology and machine learning. The paper presents a comprehensive study that combines data collection, image analysis, and machine learning techniques to classify tongue images into disease categories. The goal is to assess the feasibility and accuracy of utilizing tongue classification as a diagnostic tool and to explore its implications in healthcare.

This interdisciplinary approach holds promise for early disease detection, personalized healthcare, and more patient-centric medical practices. As we navigate the intricate interplay between ancient wisdom and contemporary technology, we aim to shed light on the transformative potential of tongue classification in the landscape of modern medicine.

II. PRIOR WORK

Previous research has demonstrated the effectiveness of CaffeNet in disease prediction using various medical images, such as X-rays, MRIs, and CT scans. The network's ability to learn complex patterns and subtle variations within images makes it a compelling candidate for the task of disease prediction based on tongue classification. By incorporating CaffeNet into our methodology, we aim to harness its image analysis capabilities for accurate disease classification.

CaffeNet, a deep convolutional neural network architecture, has been widely employed in various image analysis tasks, including medical image analysis. Researchers have leveraged its capabilities to extract and interpret intricate patterns and features from medical images.

In the context of this study, CaffeNet can be considered as a potential tool for automating the feature extraction and classification stages of tongue image analysis.

In response to the guidance provided by Traditional Chinese Medicine (TCM) experts, the research was structured around three distinct sets of experiments, each focusing on different criteria for tongue color classification. These criteria led to the creation of three separate classification tasks, specifically involving 6-classification, 5-classification, and 4-classification of tongue colors. These experiments aimed to address the complexities inherent in tongue classification, where different criteria and categorization standards may be employed by TCM practitioners.

The objective of these experiments was to gain insights into the computational challenges involved in recognizing certain tongue colors, shedding light on the capability of the deep learning model to distinguish between distinct tongue color categories. These experiments, therefore, form a critical component of our investigation into the feasibility and reliability of using tongue classification as a tool for disease prediction and early diagnosis.

V. RESULTS AND FUTURE SCOPE

Figure 5 presents a comprehensive overview of our classification accuracy results, detailing the performance across various tongue color classification tasks, including 6-classification, 5-classification, and 4-classification. A discernible trend emerges from our findings, demonstrating that, in general, as the number of categories within the classification tasks increases, the classification accuracy tends to decrease. Our study emphasizes the unique challenges encountered in distinguishing between specific tongue colors, particularly noting the difficulties in classifying light white, dark, and light red colors. To address these challenges, it is suggested that the expansion of our dataset size may offer a potential remedy.

Moreover, Figure 6 delves into the critical role of dataset size in the accuracy of tongue color classification. While it underscores the significance of dataset size, it also uncovers an intriguing revelation. As the dataset size expands, our research notes a decrease in classification accuracy, particularly within the same class. This suggests that, as the dataset grows, the classification accuracy may decrease within specific color categories, underscoring the need for a more extensive and diverse tongue image database to enhance accuracy further.

The study reaffirms the invaluable impact of data augmentation on our results, highlighting its substantial role in enhancing our model's performance. This augmentation of our dataset contributed to increased diversity and the quantity of training data, which in turn played a crucial role in achieving robust classification results. Our findings also underscore the effectiveness of the network modifications made to facilitate tongue color classification. These outcomes collectively offer profound insights into the challenges and potential solutions for disease prediction using tongue classification, underlining the importance of dataset size and data augmentation in achieving accurate and reliable results.

Expanding the dataset to include a more diverse range of tongue images, collaborating closely with Traditional Chinese Medicine experts to refine classification criteria, and integrating explainable AI techniques for transparency and interpretability are avenues for further exploration. Additionally, the potential for real-time diagnosis, the incorporation of additional patient data, clinical validation, and addressing ethical considerations stand as significant directions for advancing the practicality and impact of tongue classification in disease prediction. The research provides a strong foundation for future endeavors in this interdisciplinary field, with the goal of improving healthcare diagnostics and patient care.

Conclusion

In conclusion, our research demonstrates the potential of tongue classification in disease prediction. We find that as the number of classification categories increases, accuracy tends to decrease. The impact of dataset size, data augmentation, and network modifications on accuracy is evident. Future work should expand the dataset, collaborate across disciplines, and integrate explainable AI techniques. These outcomes lay the foundation for bridging traditional and modern diagnostic methods to enhance patient-centric healthcare with more accurate disease prediction solutions.

References

[1] Jia Y, Shelhamer E, Donahue J, et al. Caffe: Convolutional Architecture for Fast Feature Embedding[C]. acm multimedia, 2014:675-678. [2] CHEN Song-he. Study on tongue color analysis of digital tongue [D]. Beijing University of Chinese Medicine, 2007. [3] NI Hao. Study on Color - and Texture - Based Retrieval Technology of Chinese [D]. Guangdong University of Technology, 2011. [4] WU Xia. Pattern Recognition of Tongue Colors [D]. Nankai University, 2007. [5] G.Montavon,M. L. Braun, and K.-R.Mjuller, ?Kernel analysis of deep networks,? The Journal of Machine Learning Research, vol.12, pp. 2563?2581, 2011. [6] Li, J., Yuan, P., Hu, X., Huang, J., Cui, L., Cui, J., ... & Xu, J. (2021). A tongue features fusion approach to predicting D prediabetes and diabetes with machine learning. Journal of biomedical informatics, 115, 103693. [7] Naveed, S. (2022). Early diabetes discovery from tongue images. The Computer Journal, 65(2), 237-250. [8] Wan, H., Wang, Y., Fang, S., Chen, Y., Zhang, W., Xia, F., ... & Lu, Y. (2020). Associations between the neutrophil-to-lymphocyte ratio and diabetic complications in adults with diabetes: a cross-sectional study. Journal of diabetes research, 2020. [9] Hsu, P. C., Wu, H. K., Huang, Y. C., Chang, H. H., Lee, T. C., Chen, Y. P., ... & Lo, L. C. (2019). The tongue features are associated with type 2 diabetes mellitus. Medicine, 98(19). [10] Ilyas, H., Ali, S., Ponum, M., Hasan, O., Mahmood, M. T., Iftikhar, M., & Malik, M. H. (2021). Chronic kidney disease diagnosis using decision tree algorithms. BMC nephrology, 22(1), 1-11. [11] G. Maciocia, The foundations of Chinese medicine, Churchill Livingstone, London, 1989. [12] M. Porker, Theoretical foundations of Chinese medicine: systems of correspondence, MIT Press, Cambridge, 1978. [13] L. Bridges, Face reading in Chinese medicine, Churchill Livingstone, 2004. [14] B. Liu, and T. Wang, Inspection of face and body for diagnosis of diseases, Foreign Languages Press, 2002. [15] Shi, Dan, et al. \"An annotated dataset of tongue images supporting geriatric disease diagnosis.\" Data in brief 32 (2020): 106153. [16] Xu, Qiang, et al. \"Multi-task joint learning model for segmenting and classifying tongue images using a deep neural network.\" IEEE journal of biomedical and health informatics 24.9 (2020): 2481-2489.

Copyright

Copyright © 2023 Himanshu Kale. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.