Premium
A deep feature fusion methodology for breast cancer diagnosis demonstrated on three imaging modality datasets
Author(s) -
Antropova Natalia,
Huynh Benjamin Q.,
Giger Maryellen L.
Publication year - 2017
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1002/mp.12453
Subject(s) - cad , artificial intelligence , modality (human–computer interaction) , computer science , preprocessor , mammography , convolutional neural network , computer aided diagnosis , pattern recognition (psychology) , digital mammography , feature (linguistics) , deep learning , breast cancer , breast imaging , breast ultrasound , feature extraction , medical imaging , medicine , cancer , philosophy , engineering drawing , engineering , linguistics
Background Deep learning methods for radiomics/computer‐aided diagnosis ( CAD x) are often prohibited by small datasets, long computation time, and the need for extensive image preprocessing. Aims We aim to develop a breast CAD x methodology that addresses the aforementioned issues by exploiting the efficiency of pre‐trained convolutional neural networks ( CNN s) and using pre‐existing handcrafted CAD x features. Materials & Methods We present a methodology that extracts and pools low‐ to mid‐level features using a pretrained CNN and fuses them with handcrafted radiomic features computed using conventional CAD x methods. Our methodology is tested on three different clinical imaging modalities (dynamic contrast enhanced‐ MRI [690 cases], full‐field digital mammography [245 cases], and ultrasound [1125 cases]). Results From ROC analysis, our fusion‐based method demonstrates, on all three imaging modalities, statistically significant improvements in terms of AUC as compared to previous breast cancer CAD x methods in the task of distinguishing between malignant and benign lesions. ( DCE ‐ MRI [ AUC = 0.89 (se = 0.01)], FFDM [ AUC = 0.86 (se = 0.01)], and ultrasound [ AUC = 0.90 (se = 0.01)]). Discussion/Conclusion We proposed a novel breast CAD x methodology that can be used to more effectively characterize breast lesions in comparison to existing methods. Furthermore, our proposed methodology is computationally efficient and circumvents the need for image preprocessing.