Text
KLASIFIKASI LESI PRA-KANKER SERVIKS MELALUI CITRA MEDIS IVA MENGGUNAKAN METODE CONVOLUTIONAL NEURAL NETWORK DAN CLASS ACTIVATION MAPPING
The high-level infection of Human Papilloma Virus (HPV) causes cervical cancer to become the second most common disease that affects women, especially in low-middle income countries (LMICs), including Indonesia. Early screening for pre-cancerous lesions is one way to reduce the spread of cervical cancer. Visual inspection with acetic acid (VIA) is one initial screening methods recommended by the World Health Organization (WHO) to be applied in LMICs. However, healthcare professional is needed to read the VIA test diagnosis. This is because the anatomy of cervical pre-cancerous lesions is so complex. The lack healthcare professional is a problem in implementing the VIA test, especially in Indonesia. Therefore, artificial intelligence technology that is able to predict cervical pre-cancerous lesions automatically and accurately needs to be developed. This study developed a classification model for cervical pre-cancerous lesions on VIA medical images using the Convolutional Neural Network (CNN) method. The three CNN architectures used include the VGG16, VGG19 and ResNet50 architectures. Because the amount of data used is still limited, two test cases were carried out, the first without data augmentation and the second with data augmentation. The results obtained show that the performance of model classification ResNet50 pre-train with data augmentation outperformed other classification models with a performance value 91.28% accuracy, 98.75% precision, 86.81% sensitivity, 92.402% f1-score and 98.27% specificity. In addition, to increase the confidence of the CNN model results, which have been considered a black box, the interpretation of the CNN model results in heatmap visualization to show the area generated by CNN is shown. The interpretation methods used are Class Activation Mapping (CAM) and Gradient-Weight Class Activation Mapping (Grad-CAM). The visualization results obtained show that the model can be used by healthcare professionals as an alternative to the automatic initial screening of the VIA test. Keywords: Classification, Localization, CNN, CAM, Grad-CAM, Visual inspection with acetic acid.
Inventory Code | Barcode | Call Number | Location | Status |
---|---|---|---|---|
2207002579 | T74742 | T747422022 | Central Library (Referens) | Available but not for loan - Not for Loan |
No other version available