Hybrid CNN-LSTM Networks for Enhanced Lung Cancer Detection and Classification from CT Images

In this paper, a hybrid deep learning model based on Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks are introduced for the automated detection of lung cancer from CT scans. With the hope of improving the accuracy and robustness of lung cancer diagnosis, the hybrid model adopts CNNs to extract spatial features for each CT slice and LSTMs for discovering temporal dependencies between adjacent sequential slices. An overall accuracy of 92 %, higher than both CNN-only (88 %) and LSTM-only (85%) models, is achieved by evaluating the model with a labeled dataset of CT scans. In addition, the hybrid model also showed 94 % precision, i.e., fewer false positives, as well as 9 1 % recall, avoiding false negatives, which is highly important for early detection. Moreover, the F1-score of 9 2 % further reinforces the fact that the model is well-balanced in detecting cancerous lesions and avoiding errors. Furthermore, an effectiveness comparison with state-of-the-art approaches like Setio et al. (2016) and Liu et al. (2020), which were able to obtain abstract levels of accuracy of 87 % and 85 %, respectively, is presented to demonstrate a significant advance of the diagnostic performance. This model performed very well according to ROC curve analysis with an AUC of 0.97, which confirmed that it can distinguish between cancerous and noncancerous scans. The results show that not only does the hybrid CNN-LSTM model improve lung cancer detection, but it also serves as a robust diagnostic aiding tool in an automated manner. Further optimization of this approach could improve clinical decisionmaking, decrease radiologist dependence, and increase earlystage cancer detection, resulting in better patient outcomes.