Enhanced ConvNeXt-V2 based model for Accurate Lymph Node Metastasis Prediction in Breast Cancer Histopathology

Abstract

Breast cancer is a leading cause of cancer-related deaths worldwide. Delayed or missed detection of lymph node metastasis can greatly affect patient’s condition. Early identification of metastatic spread is vital for effective treatment planning and patient survival. To address the limitations of manual examination, such as high time consumption and expertise requirements, deep learning-based approaches are employed. This study presents a deep learning framework for predicting metastasis using Whole Slide Images (WSIs) from the PatchCamelyon (PCam) dataset. This work aims to increase the area under the ROC curve (AUC) while lowering false- negative predictions. To a E-mail: chieve this implemented three strong convolutional neural network models ResNet-50, EfficientNet-B3, and DenseNet-121 with the Convolutional Block Attention Module (CBAM). To address the limitations of explicit attention designs, proposed Enhanced ConvNeXt-V2 based model, a powerful fully convolutional architecture. Classification head is modified to the ConvNext V2 model. This model features large kernel depth wise convolutions and Global Response Normalization, effectively capturing local features and the overall context in the constrained 96 × 96 histopathology patches. Results indicate that proposed model surpassed the performance of all the models that used attention augmentation. On the test set, it scored 0.9710 in terms of AUC and 0.9116 in terms of F1, and at the same time, it largely lowered the number of false negatives. Gradient weighted Class Activation Mapping (Grad-CAM) analysis shows that, in comparison to attention models, proposed model is more accurate in highlighting areas of malignant cell clusters and does so with less background noise. The results indicate that modern convolutional techniques can effectively replace explicit attention mechanisms for cancer cell detection, offering a strong solution for the prediction of breast cancer metastasis.