In a recent article published in Scientific Reports, researchers introduced a novel model called "Histopathological Imaging for the Early Detection of Lung and Colon Cancer via Ensemble Deep Learning (HIELCC-EDL)." This model was designed to improve the early detection of lung and colon cancer by leveraging advanced imaging techniques. The goal was to develop an efficient method for accurately identifying cancerous tissues from images, thereby enhancing early diagnosis and improving patient outcomes through timely interventions.
Image Credit: Komsan Loonprom/Shutterstock.com
Background
The World Health Organization (WHO) identifies cancer as one of the leading causes of death worldwide, largely due to its diverse forms, aggressive nature, and high potential for metastasis. Lung and colon cancers (LCCs) are among the most prevalent, affecting both men and women across the globe. Early and accurate detection is critical to improving survival rates and guiding effective treatment plans.
Traditional methods, such as imaging and biopsies, are often time-consuming and require specialized expertise. However, recent advancements in artificial intelligence—particularly machine learning and deep learning—present new opportunities to automate and enhance cancer detection. These technologies can quickly and cost-effectively analyze large datasets, offering significant value in medical diagnostics.
About the Research
In this study, the authors introduced an ensemble deep learning model, HIELCC-EDL, designed for the early and accurate detection of lung and colon cancer (LCC) using histopathological imaging. The model was trained on a dataset of 25,000 images, which were generated through augmentation from 750 original LCC tissue images. These images were divided into five categories: lung benign tissue, lung adenocarcinoma, lung squamous cell carcinoma, colon adenocarcinoma, and colon benign tissue.
To enhance image quality, the Wiener filtering (WF) method was applied to remove noise and improve clarity, a critical preprocessing step for ensuring accurate model performance. For feature extraction, the channel attention Residual Network (CA-ResNet50) model was employed to capture complex patterns in the images. The hyperparameters of CA-ResNet50 were optimized using the tuna swarm optimization (TSO) technique, inspired by the hunting behavior of tuna fish.
The detection of LCC was achieved through an ensemble of three classifiers: convolutional neural networks (CNNs), extreme learning machine (ELM), and long short-term memory (LSTM). This ensemble approach combined the strengths of each classifier, resulting in more robust and precise recognition of cancerous tissues. The model’s performance was evaluated using several metrics, including accuracy, precision, recall, F1 score (harmonic mean of precision and recall), and area under the curve (AUC) score.
Research Findings
The results showed that the proposed HIELCC-EDL model outperformed existing models across all key performance metrics, achieving an average accuracy of 99.60 %, precision of 99.00 %, recall of 99.00 %, F1 score of 99.00 %, and an AUC score of 99.38 %. The study also confirmed that the TSO technique effectively optimized the model’s hyperparameters, contributing to its high performance.
An analysis of feature importance demonstrated that the CA-ResNet50 model successfully captured complex patterns, significantly enhancing the accuracy of cancer detection. The model also proved robust against various types of noise and perturbations, further highlighting its reliability.
To ensure model interpretability, the researchers employed techniques such as saliency maps and feature importance analysis to examine how the model made its predictions. The findings revealed that the model accurately identified critical features of cancerous tissues, including morphological changes and textural patterns. These features aligned with the known biological characteristics of lung and colon cancers, reinforcing the model’s potential as a reliable diagnostic tool.
Applications
The newly introduced technique holds significant promise for the early detection and diagnosis of LCCs). By automating the identification of cancerous tissues from histopathological images, it can assist pathologists in making more accurate and timely diagnoses, leading to earlier interventions, improved treatment outcomes, and higher survival rates. Its exceptional accuracy and efficiency also make it a valuable tool for large-scale cancer screening programs.
Additionally, the model's adaptability allows it to be applied to other medical imaging fields, such as the diagnosis of breast and prostate cancers, with appropriate retraining on new datasets. Beyond healthcare, its robust capabilities can be extended to fields like computer vision and image processing, broadening its potential applications.
Conclusion
In summary, the HIELCC-EDL model demonstrated high effectiveness in accurately detecting LCCs from histopathological images with remarkable precision. The study emphasized the value of ensemble deep learning models in improving early cancer diagnosis and treatment outcomes while showcasing the model’s adaptability for other medical imaging applications.
Future research should focus on extending the model to detect other cancers, enhancing its performance with larger datasets and advanced optimization techniques. Additionally, further exploration of the model's interpretability is necessary to better understand its prediction processes and identify any potential biases. Overall, this study highlights the importance of ongoing innovation in machine learning and deep learning for medical applications, with the potential to revolutionize disease detection, including cancer.
Journal Reference
Alotaibi, M., Alshardan, A., Maashi, M.et al. Exploiting histopathological imaging for early detection of lung and colon cancer via ensemble deep learning model. Sci Rep 14, 20434 (2024). DOI: 10.1038/s41598-024-71302-9, https://www.nature.com/articles/s41598-024-71302-9
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.