The cervical screening exam is a widely used method to detect cervical cancer and precancerous lesions. Automated classification of the results can assist healthcare professionals in accurately identifying abnormal cytology patterns, increasing accuracy and consistency in detecting anomalies. Furthermore, systematizing this solution can reduce analysis time and associated costs, enabling the provision of an immediate pre-diagnosis,
especially in remote areas. This approach also has the potential for integration into public health systems, contributing to more efficient and accessible care. Therefore, this study proposes the application of pre-trained convolutional neural network models VGG16 and VGG19 for classifying images resulting from the liquid-based cytology technique, comparing the performance of 4-class versus 2-class classification with balanced and unbalanced data. Several architectures were tested, and accuracies of up to 98% were achieved, along with
good classification metrics, showing potential as a solution to assist healthcare professionals in more assertive classification of these results.
|
