Multimodal Lung Cancer Diagnosis: Combining CT Image Features with Metadata Using CNN and RF

Lung cancer is one of the leading causes of cancer deaths it is, therefore, important to have accurate and timely diagnosis tools. This paper gives a dual approach through which we shall employ CNNs for feature extraction on the CT images of patients while incorporating the secondary structured metadata on the patient such as age and smoking history through the application of RF. The findings also reveal that the proposed hybrid model offers a higher accuracy than individual CNN and RF models. When validating the analyzed output of a publicly available dataset, the effectiveness of the proposed hybrid model was observed to be 93.1% better than CNN (89.5%) and RF (85.3%). The performance measure of precision and recall was 0.910 and 0.924 respectively and thus the ability to minimize the false positive and false negative cases, which is more important in the context of early cancer diagnosis. The F1-score of 91.7%, shows that the model is equally good at both precision and recall, although that the AUC-ROC of 0.95 shows its discriminative capability is slightly better t CNN: 0.92 and RF: 0.88. For the malignant instances, the F1-scores calculated were 0.91, while for benign cases, the F1-scores obtained were 0.92 which supports the evidence of this model's robust state irrespective of the classifications. The feature importance analysis revealed that image features extracted by CNN accounted for about 70% of the total classification accuracy while 30% belonged to metadata with different modalities emphasizing the importance of the fusion of the multimodal data.. The studies thus establish the hybrid model for a dependable, interpretable, and scalable diagnostic tool for lung cancer diagnosis. The future studies might expand this approach to other disorders and evaluating the applicability of the method to real-world clinical settings.