Download PDF
ABSTRACT
Cervical cancer presents a significant global health challenge, underscoring the importance of early detection to prevent and manage the disease. This study utilizes a dataset comprising 1500 JPG images (224x224x3) of cervical precancerous lesions, sourced from the IARC cervical cancer image bank. The images are classified as positive, negative, or suspicious for cancer. Employing Python, TensorFlow, Keras, OpenCV, NumPy, and Pandas, the research applies various preprocessing techniques, including data augmentation and CLAHE. The model architecture incorporates a CNN with Squeeze-and-Excitation blocks for feature refinement. The dataset is split into an 80:20 ratio, and the model undergoes 100 epochs of training with a batch size of 32 and a learning rate of 0.001. Evaluation metrics encompass accuracy, precision, recall, F-Measure, confusion matrices, and ROC curves. Preprocessing significantly boosts classification accuracy compared to a model without preprocessing. Using k-fold cross-validation (k=3) on test data before preprocessing, the model achieves an accuracy of 94.24%. While initial findings reveal promising ROC-AUC scores, erratic accuracy and loss graphs underscore the necessity for preprocessing. Misclassifications in validation and test sets are attributed to data limitations. While after preprocessing techniques, such as CLACHE, dataset expansion, and increased epochs, the accuracy is elevated to 100%, with ROC-AUC scores reaching 100% for all classes. The confusion matrix indicates no misclassifications post-preprocessing, while the classification report showcases optimal precision, recall, F1-score, and accuracy across datasets, highlighting the efficacy of preprocessing techniques. This study underscores the pivotal role of preprocessing techniques in enhancing the accuracy of detecting cervical precancerous cells. It emphasizes their significance in early lesion identification and cervical cancer risk screening.
Keywords: Cervical Cancer: Squeeze-and-Excitation Blocks: Preprocessing Techniques