AN AUTOMATED LUNG CANCER DETECTION MODEL BASED ON SELF-GUIDED BAT AND DEEP FOREST TECHNIQUES

Medical imaging techniques like CT scans are widely used for diagnosing lung cancer and malignancy. These techniques do not require sampling and also reduce the risk of spread to lung cancer nodules. Further, advanced computer diagnostic tools can process the CT images and classify the images in tumor and non-tumor labels. These tools have a significant and crucial impact on earlier lung cancer diagnoses and also improve the survivability rate of affected patients. In the medical field, accuracy is an important parameter for evaluating the performance of any diagnostic tool, and it can be on the higher side. Hence, this work proposes an automated lung cancer diagnostic model, SgBA–iForest, with a higher accuracy rate. This work presents the two algorithms based on the self-guided bat algorithm and isolation forest for detecting lung nodules. These algorithms are integrated into the proposed detection model. The multi-level thresholding method is utilized for the segmentation task and is responsible for multi-level thresholding to obtain the region of interest. Several features in the context of statistical, texture, geometrical and shape, etc., are extracted from the segmented area. The irrelevant features are identified based on the self-guided bat algorithm (SgBA). It is an improved variant of the bat algorithm that can address the local optima and tradeoff between loudness and pulse rate parameters issues of the bat algorithm using updated search mechanisms. Finally, an isolation Forest (iForest)-based technique is adopted to detect lung cancer. The performance of the proposed detection model, SgBA–iForest is validated using the Lung CT-Diagnosis database and analyzed using accuracy, precision, sensitivity, specificity, F1-score and ROC parameters based on 10-cross-fold validation. The simulation results indicated that the proposed SgBA–iForest achieves higher accuracy (97.56%), recall (98.11%), F-score (98.43%), kappa (96.74%) and ROC (97.28%) results.