Papillary Thyroid Cancer Histopathological Image Classification Using Pretrained ConvNeXt Tiny and Grad-CAM Interpretation

The use of histopathological images for the diagnosis of all types of cancer, including thyroid cancer, is considered the gold standard in clinical practice. Even so, the process of manually diagnosing histopathological images remains a challenge because this diagnosis process takes a long time and has problems in terms of inconsistencies and disagreements between experts. The development of computer-aided technology utilizing deep learning has enabled the implementation of a system for identifying and classifying thyroid cancers based on histopathological images. Despite several studies having been carried out on thyroid cancer classification using deep learning, limited model architectures have been evaluated. Moreover, model interpretability, which is critical for its clinical acceptance, remains underexplored. to expand current research on Papillary Thyroid Cancer (PTC) classification, this study implemented ConvNeXt Tiny, a new generation of convolutional networks, to classify PTC-like and non-PTC-like histopathological images. The Grad-CAM technique was used to address the lack of interpretability in previous research. The current study contributes to the field of PTC histopathological image analysis by combining a CNN-based model and Grad-CAM for both classification and interpretation purposes. Given the absence of advanced preprocessing, the accuracy achieved was approximately 84.36%. This suggests that the implemented model has potential for further development into a more robust version. Visualization and interpretation of the model results were performed using Grad-CAM in the form of a class-activation map.