Hybrid Learning in Medical Imaging: CNN-SVM Applications in Lung Cancer Detection

Despite progress achieved in the ongoing fight against cancer, lung cancer remains one of the leading causes of cancer death globally, which substantiates the need for rapid and accurate diagnostic methods. This research presents a heterogeneous model that uses CNNs for feature extraction and SVMs for classification. The LIDC-IDRI dataset was used to assess the model. It yielded an 89.7% accuracy, far beyond the performance of other sole algorithms like CNN at 85.2%, SVM at 78.9%, Random forest at 82.3% and Logistic regression at 75.6%. The accuracy of the hybrid model was an F1-Score of 87.9%, precision of 88.5%, and recall of 87.4%, with an AUC of 91.2%, clearly defining the models' efficiency for classifying between malignant and benign cases. Feature map representations proved that CNN could learn hierarchical features, from basic texture to complex nodular structures, which empowered the SVM classifier to achieve higher accuracy and generalization. The training and validation curves showed progressive learning and no sign of overfitting, as the validation accuracy reached 92%. Tuning of variables improved the SVM classifier as it remained at its best of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{C}=\mathbf{10}$</tex> while <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\gamma=0.1$</tex>. The study's results demonstrate the effectiveness of the CNN-SVM combination in improving early lung cancer diagnosis and increasing diagnostic accuracy without compromising computational capacity. The present work establishes the foundation for integrating advanced deep learning models in clinical environments to promote improved patient prognosis through early diagnosis and personalized treatment frameworks.