AI-Powered Diagnosis: A CNN-SVM Framework for Identifying COVID-19 in Chest Radiographs

The current COVID-19 outbreak has clearly shown that accurate, rapid and sensitive diagnostic tools are need of the hour. Despite the fact that RT-PCR assays remain an acknowledged standard, the observed limitations of these methods make further diagnostic tests desirable. This research proposes a CNN-SVM model for the CXRs classification as COVID-19 positive or negative combining the CNN feature extraction and the exhaustive classification performance of SVMS. The efficiency of the proposed framework has been evaluated using the public available CXR dataset, which was made augmented to address class imbalance issue. The cumulative model CNN-SVM achieved the best performance of 93.2% which is much higher than that of the separate CNN and SVM having achieved 89.5% and 85.2% accuracies respectively. The hybrid model exhibited enhanced precision (92.5%), recall (93.8%), and F1-score (93.1%) relative to the CNN-only model (precision: 88. Precision: 7%, recall: 90.2 %; F1 score: 89.4%. The AUC-ROC value for the proposed framework was 0.96, which illustrated that the COVID-19 positive and negative cases can be differentiated more accurately using the proposed framework than using only the CNN or SVM techniques, with an AUC-ROC of 0.92 for CNN and 0.88 for SVM. The training and validation curves of the losses gave a clear conviction of efficient learning with less overfitting, while the confusion matrix yielded 85% accuracy in COVID-positive cases and less misclassification. These results support the reliability and validity of the hybrid model when used in clinical contexts. Future work will focus on increasing the variety of datasets and finding ways to enhance pre-training approaches. The study leads to understanding CNN-SVM as efficient tool to apply artificial intelligence in COVID-19 diagnosis.