Abstract:
By using state of the art deep learning models on the Iraq Oncology Teaching Hospital/National
Center for Cancer Diseases (IQ-OTH/NCCD) dataset, a novel advancement in lung cancer
prediction is demonstrated in this study. Our analysis shows the Compact Convolutional
Transformers (CCT) to be the clear choice among five cutting edge models, with an incredible
accuracy of 99.09%. Building on this achievement, we carried out an in depth ablation study to
further optimize the CCT. The effects of optimizers, learning rates, loss functions, batch sizes,
and pooling techniques were examined in detail in this study. A careful adjustment of these
parameters produced a notable improvement in accuracy, highlighting the crucial part that fine
tuning performs in building predictive models. Further, we conducted a thorough investigation
using significant metrics such confusion matrices, classification reports, Area Under the Curve
(AUC) scores, and loss curves to verify the robustness of our method. The model performed
quite well, classifying cases properly and providing detailed insights into its recall and precision.
The most significant conclusion of our research is that our best model reaches an astounding
accuracy of 99.09%, highlighting its potential as an effective tool for early lung cancer
identification. This achievement highlights the value of using deep learning in medical
diagnostics in addition to marking a significant improvement in predicted accuracy.
