Abstract:Background: Cervical cancer is the fourth most common gynecological malignancy in the world. Traditional cervical cancer cell screening usually requires a pathologist to observe thousands of cells under microscope and give a report against the diagnostic criteria. This method is not only time-consuming and labor-intensive, but also relies heavily on the doctor's experience and has a strong subjectivity. Cervical cancer cells detection is an essential method to prevent cervical cancer. The accuracy of traditional computer-aided cancer cell detection is low, because of cells overlapping. Methods: In this paper, we propose a detection method based on multi-scale feature fusion (DC-ResNet). Secondly, we introduce dilated convolution and group convolution in the backbone with deep residual network to improve resolution and model expression ability and obtain better cell feature representation. Multi-scale features are generated based on feature pyramid fusion network to ensure the detection of small cells. Then, based on unsupervised clustering, adaptive anchors of cervical cancer cells are generated to further improve the detection rate. In addition, the unbalanced sample contribution loss is reduced by defining a new Balanced L1-based loss function. Findings: Our algorithm adopts an end-to-end deep learning method to realize the detection of cancer cells on multi-cells pictures, and obtain the location while identifying cancer cells. The experiments results show that the proposed method achieves 50.4% mAP, which is 3 percentage points higher than the traditional method. Interpretation: DC-ResNet, a deep neural network based on multi-scale feature fusion to detecte cervical cancer cell, demonstrated good detection performance.Funding: National Natural Science Foundation of China (No.61472042, No.61802020); and the Beijing Natural Science Foundation (No.4174094).Declaration of Interest: None to declare.

Multi-view Learning with Feature Level Fusion for Cervical Dysplasia Diagnosis

Multi-Modal Fusion Learning for Cervical Dysplasia Diagnosis.

Deep Fusion Network Based Sparse View CT Reconstructions for Clinical Diagnostic Scanners

Multi-model Data Fusion for Cervical Precancerous Lesions Detection

Deep Learning-Based Multimodal Image Analysis for Cervical Cancer Detection.

Multi-source Data Fusion for Recognition of Cervical Precancerous Lesions

A Corresponding Region Fusion Framework for Multi-Modal Cervical Lesion Detection

Multi-state colposcopy image fusion for cervical precancerous lesion diagnosis using BF-CNN

A two-stream decision fusion network for cervical pap-smear image classification tasks

Classification of Cervical Lesions Based on Multimodal Features Fusion

Computer-Aided Cervical Cancer Diagnosis Using Time-Lapsed Colposcopic Images

Deep integrated fusion of local and global features for cervical cell classification

Prototypical Cross-domain Knowledge Transfer for Cervical Dysplasia Visual Inspection

DeepCervix: A deep learning-based framework for the classification of cervical cells using hybrid deep feature fusion techniques

Interpretation on Deep Multimodal Fusion for Diagnostic Classification

FMA-Net: Fusion of Multi-Scale Attention for Grading Cervical Precancerous Lesions

CTIFI: Clinical-experience-guided three-vision images features integration for diagnosis of cervical lesions

Cervical Cancer Cell Detection Based on Multi-Scale Feature Fusion

Interpretable Detector for Cervical Cytology Using Self-Attention and Cell Origin Group Guidance

Towards Efficient Information Fusion: Concentric Dual Fusion Attention Based Multiple Instance Learning for Whole Slide Images

CIRF: Coupled Image Reconstruction and Fusion Strategy for Deep Learning Based Multi-Modal Image Fusion