Artificial Intelligence in Cancer Detection: A Neural Network Approach to Differentiating Malignant and Benign Cells
Anikait Sota · 2025
This paper explores the application of artificial intelligence in the diagnosis of cancer, specifically in making a distinction between malignant and benign cells based on neural network models. Traditional diagnostic methods include biopsies and imaging, which are generally invasive, time-consuming, and expensive. A dataset from the University of Wisconsin was applied to train and test two machine learning models: a custom neural network and a Multi-Layer Perceptron (MLP) classifier implemented in scikit-learn. The Sigmoid-Relu-Relu-Sigmoid custom neural network attained an accuracy of 92.11% with an F1 score of 0.91 and an AUC of 0.94, thereby showing a good tradeoff between accuracy and generalization. By contrast, the MLP classifier, trained on a subset of top predictive features, achieved a comparable accuracy of 92.0% with an F1 score of 0.88 and an AUC of 0.90, providing a computationally friendly alternative. Analysis revealed that features representing extreme tumor characteristics, such as radius_worst and texture_worst, contributed significantly to model performance, underscoring the importance of capturing aggressive tumor properties in cancer diagnosis.