Deep Convolutional Neural Networks (D-CNNs) for Breast Cancer Detection

OVERFIT: AI, Machine Learning, and Deep Learning Made Simple

Here we discuss three different papers (see links below) on using D-CNNs to detect breast cancer.

The first source details the development and evaluation of HIPPO, a novel explainable AI method that enhances the interpretability and trustworthiness of ABMIL models in computational pathology. HIPPO aims to address the challenges of opaque decision-making in D-CNNs by generating counterfactual examples through tissue patch modifications in whole slide images. This allows for a deeper understanding of model behavior and the identification of potential biases. The second source investigates the performance of various D-CNN architectures, including transfer learning and an ensemble model, in breast cancer detection. This study finds that an ensemble model provides the highest detection and classification accuracy, while transfer learning does not improve the performance of the original D-CNN models. The authors attribute this to potential negative transfer learning, where pre-trained models trained on large-scale datasets with natural images may not be suitable for microscopic images or images from a different domain. The study concludes that the ensemble model, termed 'DIR', demonstrates promising results in breast cancer detection and highlights the potential for future research to address limitations and further enhance the accuracy of D-CNNs for this application.

Read:

• A study on Deep Convolutional Neural Networks, Transfer Learning and Ensemble Model for Breast Cancer Detection
• Explainable AI for computational pathology identifies model limitations and tissue biomarkers
• Crown-Like Structures in Breast Adipose Tissue: Finding a 'Needle-in-a-Haystack' using Artificial Intelligence and Collaborative Active Learning on the Web