Abstract:PurposeImage‐based breast lesion detection is a powerful clinical diagnosis technology. In recent years, deep learning architectures have achieved considerable success in medical image analysis however, they always require large‐scale samples. In mammography images, breast lesions are inconspicuous, multiscale, and have blurred edges. Moreover, few well‐labeled images exist. Because of these factors, the detection accuracy of conventional deep learning methods is low. Therefore, we attempted to improve the accuracy of mammary lesion detection by introducing transfer learning (TL) into a deep learning framework for the few‐shot learning task and thus provide a method that will further assist physicians in detecting breast lesions.MethodsIn this paper, we propose a method called “few‐shot learning with deformable convolution for multiscale lesion detection in mammography,” named FDMNet. Deformable convolution is introduced for enhancing the network’s ability to detect lesions, and the sensitivity of the multiscale feature space is reinforced by using a feature pyramid method. Furthermore, by introducing location information in the predictor, the sensitivity of the model to lesion location is also enhanced. The proposed method, through the TL technique that is applied mines the potentially common knowledge of features in the source domain and transfers it into the target domain to improve the accuracy of breast lesion detection in the few‐shot learning task.ResultsOn the publicly available datasets for screening mammography CBIS‐DDSM and Mini‐MIAS, the proposed method performs better than five widely used detection methods. On the CBIS‐DDSM dataset, its comprehensive scores, sensitivity, precision, and the mean dice similarity coefficient are 0.911, 0.949, 0.873, and 0.913, respectively, and on the Mini‐MIAS dataset, these values are 0.931, 0.966, 0.882, and 0.941, respectively.ConclusionsTo achieve the few‐shot learning required for medical image analysis, the proposed method uses TL to execute feature knowledge transformation and includes deformable convolution to build a feature pyramid structure, which enhances the learning performance of the network for lesions. The results of comparative numerical experiments show that the proposed method outperforms some state‐of‐the‐art methods.

Few-Shot Tumor Detection via Feature Reweighting and Knowledge Transferring

Medical tumor image classification based on Few-shot learning

CONSULT: Contrastive Self-Supervised Learning for Few-shot Tumor Detection

Whole-slide-imaging Cancer Metastases Detection and Localization with Limited Tumorous Data

Few-Shot Breast Cancer Metastases Classification via Unsupervised Cell Ranking

TumorDet: A Breast Tumor Detection Model Based on Transfer Learning and ShuffleNet

A Novel Marker-Less Lung Tumor Localization Strategy on Low-Rank Fluoroscopic Images with Similarity Learning

Few-shot Learning with Deformable Convolution for Multi-Scale Lesion Detection in Mammography.

Robust and accurate pulmonary nodule detection with self-supervised feature learning on domain adaptation

Acquiring Weak Annotations for Tumor Localization in Temporal and Volumetric Data

Multi-Resolution CNN and Knowledge Transfer for Candidate Classification in Lung Nodule Detection

Improving segmentation and classification of renal tumors in small sample 3D CT images using transfer learning with convolutional neural networks

Lesion Feature Extraction and Classification Optimization Method Using Dynamic Fusion of Global Attention and Local Attention

A Hybrid Deep Learning Model for Breast Cancer Detection and Classification

Pan-tumor T-lymphocyte detection using deep neural networks: Recommendations for transfer learning in immunohistochemistry

Inception UNet architecture for breast tumor segmentation and detection using hybrid deep learning approach

Enhancing Breast Cancer Classification: A Few-Shot Meta-Learning Framework with DenseNet-121 for Improved Diagnosis

Cancer detection for small-size and ambiguous tumors based on semantic FPN and transformer

DBTN: An adaptive neural network for multiple-disease detection via imbalanced medical images distribution

Boosting Medical Image-based Cancer Detection via Text-guided Supervision from Reports

Tumor Detection Using Deep Learning Method in Automated Breast Ultrasound.