Multi-view Local Co-occurrence and Global Consistency Learning Improve Mammogram Classification Generalisation
This work addresses the challenge of improving breast cancer screening accuracy for medical imaging by integrating multi-view information, representing an incremental advance over existing deep learning methods.
The paper tackled the problem of limited accuracy in deep learning systems for mammogram classification by proposing a multi-view global-local analysis method that mimics radiologists' reading of complementary ipsilateral views. The model outperformed competing methods in classification accuracy and generalization on a large-scale private dataset and two public datasets.
When analysing screening mammograms, radiologists can naturally process information across two ipsilateral views of each breast, namely the cranio-caudal (CC) and mediolateral-oblique (MLO) views. These multiple related images provide complementary diagnostic information and can improve the radiologist's classification accuracy. Unfortunately, most existing deep learning systems, trained with globally-labelled images, lack the ability to jointly analyse and integrate global and local information from these multiple views. By ignoring the potentially valuable information present in multiple images of a screening episode, one limits the potential accuracy of these systems. Here, we propose a new multi-view global-local analysis method that mimics the radiologist's reading procedure, based on a global consistency learning and local co-occurrence learning of ipsilateral views in mammograms. Extensive experiments show that our model outperforms competing methods, in terms of classification accuracy and generalisation, on a large-scale private dataset and two publicly available datasets, where models are exclusively trained and tested with global labels.