A Comparative Study of Classification Algorithms for Enhanced Lung Cancer Prediction Using Deep Learning and SOM-Based Microscopic Image Analysis

Not Approved

Relevance and Originality

This research article addresses a pressing health issue—lung cancer detection—by proposing a novel hybrid approach that combines deep learning with Self-Organizing Maps (SOM). The relevance is underscored by the significant global impact of lung cancer, making the exploration of advanced detection techniques crucial. The originality of the study lies in its integration of different methodologies, particularly the use of a pre-trained MobileNet model for feature extraction alongside SOM for dimensionality reduction. This innovative combination reflects a forward-thinking approach that could contribute to more precise diagnostic practices in oncology.

Methodology

The methodology presented in this study is robust and thoughtfully designed, utilizing a hybrid model that leverages both deep learning and traditional machine learning techniques. The use of MobileNet for feature extraction is particularly noteworthy, as it allows for efficient processing of high-dimensional image data. Furthermore, the application of SOM for dimensionality reduction demonstrates a sophisticated understanding of data visualization in medical imaging. However, the article could enhance clarity by detailing the dataset used, including its size, source, and preprocessing steps, as well as specifying the training and testing procedures for the machine learning classifiers employed.

Validity & Reliability

The validity of the findings is supported by the comparative analysis of various classifiers, including Random Forest, LightGBM, and Decision Tree, which allows for a thorough evaluation of the model's performance. The results showing competitive classification accuracy reinforce the reliability of the approach. However, the paper could benefit from a discussion on the evaluation metrics used to assess the classifiers, such as accuracy, sensitivity, specificity, and F1 score. Additionally, providing insights into potential biases in the data or limitations of the model would contribute to a more comprehensive understanding of the research's applicability.

Clarity and Structure

The paper is generally well-structured, guiding readers through the problem statement, methodology, and results in a logical flow. The clarity of the presentation is commendable, but some technical details could be simplified for accessibility to a broader audience. For instance, defining key terms related to deep learning and machine learning early in the article would help demystify the content for readers less familiar with these concepts. Including visual aids, such as diagrams or flowcharts illustrating the model architecture or process flow, could also enhance comprehension.

Result Analysis

The result analysis provides valuable insights into the effectiveness of the hybrid model in detecting different lung cancer subtypes. The emphasis on the promising performance of Random Forest and Decision Tree classifiers is particularly significant, suggesting practical implications for clinical applications. However, the paper would benefit from a more detailed presentation of the results, including specific performance metrics for each classifier, and visual representations of the findings, such as confusion matrices or ROC curves. Discussing the clinical relevance of the results and potential future applications of the research in real-world diagnostic settings would further strengthen the impact of the study.