Deep learning in breast imaging

The Challenge of Deep Learning for the Prevention and Automatic Diagnosis of Breast Cancer: A Systematic Review

OBJECTIVES

This review aims to evaluate several convolutional neural network (CNN) models applied to breast cancer detection, to identify and categorize CNN variants in recent studies, and to analyze their specific strengths, limitations, and challenges.

METHODS

Using PRISMA methodology, this review examines studies that focus on deep learning techniques, specifically CNN, for breast cancer detection. Inclusion criteria encompassed studies from the past five years, with duplicates and those unrelated to breast cancer excluded. A total of 62 articles from the IEEE, SCOPUS, and PubMed databases were analyzed, exploring CNN architectures and their applicability in detecting this pathology.

RESULTS

The review found that CNN models with advanced architecture and greater depth exhibit high accuracy and sensitivity in image processing and feature extraction for breast cancer detection. CNN variants that integrate transfer learning proved particularly effective, allowing the use of pre-trained models with less training data required. However, challenges include the need for large, labeled datasets and significant computational resources.

CONCLUSIONS

CNNs represent a promising tool in breast cancer detection, although future research should aim to create models that are more resource-efficient and maintain accuracy while reducing data requirements, thus improving clinical applicability.

Application of deep learning on mammographies to discriminate between low and high-risk DCIS for patient participation in active surveillance trials

BACKGROUND

Ductal Carcinoma In Situ (DCIS) can progress to invasive breast cancer, but most DCIS lesions never will. Therefore, four clinical trials (COMET, LORIS, LORETTA, AND LORD) test whether active surveillance for women with low-risk Ductal carcinoma In Situ is safe (E. S. Hwang et al., BMJ Open, 9: e026797, 2019, A. Francis et al., Eur J Cancer. 51: 2296-2303, 2015, Chizuko Kanbayashi et al. The international collaboration of active surveillance trials for low-risk DCIS (LORIS, LORD, COMET, LORETTA),  L. E. Elshof et al., Eur J Cancer, 51, 1497-510, 2015). Low-risk is defined as grade I or II DCIS. Because DCIS grade is a major eligibility criteria in these trials, it would be very helpful to assess DCIS grade on mammography, informed by grade assessed on DCIS histopathology in pre-surgery biopsies, since surgery will not be performed on a significant number of patients participating in these trials.

OBJECTIVE

To assess the performance and clinical utility of a convolutional neural network (CNN) in discriminating high-risk (grade III) DCIS and/or Invasive Breast Cancer (IBC) from low-risk (grade I/II) DCIS based on mammographic features. We explored whether the CNN could be used as a decision support tool, from excluding high-risk patients for active surveillance.

METHODS

In this single centre retrospective study, 464 patients diagnosed with DCIS based on pre-surgery biopsy between 2000 and 2014 were included. The collection of mammography images was partitioned on a patient-level into two subsets, one for training containing 80% of cases (371 cases, 681 images) and 20% (93 cases, 173 images) for testing. A deep learning model based on the U-Net CNN was trained and validated on 681 two-dimensional mammograms. Classification performance was assessed with the Area Under the Curve (AUC) receiver operating characteristic and predictive values on the test set for predicting high risk DCIS-and high-risk DCIS and/ or IBC from low-risk DCIS.

RESULTS

When classifying DCIS as high-risk, the deep learning network achieved a Positive Predictive Value (PPV) of 0.40, Negative Predictive Value (NPV) of 0.91 and an AUC of 0.72 on the test dataset. For distinguishing high-risk and/or upstaged DCIS (occult invasive breast cancer) from low-risk DCIS a PPV of 0.80, a NPV of 0.84 and an AUC of 0.76 were achieved.

CONCLUSION

For both scenarios (DCIS grade I/II vs. III, DCIS grade I/II vs. III and/or IBC) AUCs were high, 0.72 and 0.76, respectively, concluding that our convolutional neural network can discriminate low-grade from high-grade DCIS.

Low 18F-fluorodeoxyglucose dose positron emission tomography assisted by a deep-learning image-denoising technique in patients with lymphoma

Background

Patients with lymphoma receive multiple positron emission tomography/computed tomography (PET/CT) exams for monitoring of the therapeutic response. With PET imaging, a reduced level of injected fluorine-18 fluorodeoxyglucose ([18F]FDG) activity can be administered while maintaining the image quality. In this study, we investigated the efficacy of applying a deep learning (DL) denoising-technique on image quality and the quantification of metabolic parameters and Deauville score (DS) of a low [18F]FDG dose PET in patients with lymphoma.

Methods

This study retrospectively enrolled 62 patients who underwent [18F]FDG PET scans. The low-dose (LD) data were simulated by taking a 50% duration of routine-dose (RD) PET list-mode data in the reconstruction, and a U-Net-based denoising neural network was applied to improve the images of LD PET. The visual image quality score (1 = undiagnostic, 5 = excellent) and DS were assessed in all patients by nuclear radiologists. The maximum, mean, and standard deviation (SD) of the standardized uptake value (SUV) in the liver and mediastinum were measured. In addition, lesions in some patients were segmented using a fixed threshold of 2.5, and their SUV, metabolic tumor volume (MTV), and tumor lesion glycolysis (TLG) were measured. The correlation coefficient and limits of agreement between the RD and LD group were analyzed.

Results

The visual image quality of the LD group was improved compared with the RD group. The DS was similar between the RD and LD group, and the negative (DS 1-3) and positive (DS 4-5) results remained unchanged. The correlation coefficients of SUV in the liver, mediastinum, and lesions were all >0.85. The mean differences of SUVmax and SUVmean between the RD and LD groups, respectively, were 0.22 [95% confidence interval (CI): -0.19 to 0.64] and 0.02 (95% CI: -0.17 to 0.20) in the liver, 0.13 (95% CI: -0.17 to 0.42) and 0.02 (95% CI: -0.12 to 0.16) in the mediastinum, and -0.75 (95% CI: -3.42 to 1.91), and -0.13 (95% CI: -0.57 to 0.31) in lesions. The mean differences in MTV and TLG were 0.85 (95% CI: -2.27 to 3.98) and 4.06 (95% CI: -20.53 to 28.64) between the RD and LD groups.

Conclusions

The DL denoising technique enables accurate tumor assessment and quantification with LD [18F]FDG PET imaging in patients with lymphoma.

Development of a nomogram-based model combining intra- and peritumoral ultrasound radiomics with clinical features for differentiating benign from malignant in Breast Imaging Reporting and Data System category 3-5 nodules

Background

The differences in benign and malignant breast tumors are not only within the nodules but also involve changes in the surrounding tissues. Radiomics can reveal many details that are not discernible to the naked eye. This study aimed to distinguish between benign and malignant breast nodules using an ultrasound-based intra- and peritumoral radiomics model.

Methods

This study retrospectively collected the information from 379 patients with Breast Imaging Reporting and Data System (BI-RADS) category 3-5 nodules and clear pathological diagnosis of breast nodules screened by routine ultrasound examination in the Sixth People's Hospital Affiliated to Medical College of Shanghai Jiao Tong University from January 2017 to December 2022. The largest dimension of the lesion on the 2D ultrasound image was selected to outline the area of interest which was conformally and outwardly expanded automatically by 5 mm to extract intra- and peritumor radiomics features. The included cases were randomly divided into training sets and test sets in a ratio of 7:3. The optimal features of the included models were retained by statistical and machine learning methods of dimensionality reduction, and logistic regression was used as the classifier to build an intratumoral model and a combined intratumoral-peritumoral radiomics model, respectively; through single-factor and multifactor logistic regression, the optimal features that could predict benign and malignant breast tumors were screened. The clinical and imaging models were established by selecting independent risk factors as clinical and imaging features through univariate and multifactorial logistic regression.

Results

Among 379 BI-RADS category 3-5 breast nodules, there were 124 malignant nodules and 255 benign nodules; patients were aged 14 to 88 (46.22±15.51) years, and the age differences, radiomics score, and mass diameter between the training and test sets were not statistically significant (P>0.05). The intra- and peritumor radiomics model had an area under the curve (AUC) of 0.840 [95% confidence interval (CI): 0.766-0.914] in the test set. The model with intra- and peritumoral ultrasound radiomics features combined with clinical features had an AUC value of 0.960 (95% CI: 0.920-0.999).

Conclusions

The nomogram, developed using intratumoral and peritumoral radiomics features combined with clinical risk features, demonstrated superior performance in distinguishing between benign and malignant BI-RADS 3-5 lesions.

Automated Triage of Screening Breast MRI Examinations in High-Risk Women Using an Ensemble Deep Learning Model

OBJECTIVES

The aim of the study is to develop and evaluate the performance of a deep learning (DL) model to triage breast magnetic resonance imaging (MRI) findings in high-risk patients without missing any cancers.

MATERIALS AND METHODS

In this retrospective study, 16,535 consecutive contrast-enhanced MRIs performed in 8354 women from January 2013 to January 2019 were collected. From 3 New York imaging sites, 14,768 MRIs were used for the training and validation data set, and 80 randomly selected MRIs were used for a reader study test data set. From 3 New Jersey imaging sites, 1687 MRIs (1441 screening MRIs and 246 MRIs performed in recently diagnosed breast cancer patients) were used for an external validation data set. The DL model was trained to classify maximum intensity projection images as "extremely low suspicion" or "possibly suspicious." Deep learning model evaluation (workload reduction, sensitivity, specificity) was performed on the external validation data set, using a histopathology reference standard. A reader study was performed to compare DL model performance to fellowship-trained breast imaging radiologists.

RESULTS

In the external validation data set, the DL model triaged 159/1441 of screening MRIs as "extremely low suspicion" without missing a single cancer, yielding a workload reduction of 11%, a specificity of 11.5%, and a sensitivity of 100%. The model correctly triaged 246/246 (100% sensitivity) of MRIs in recently diagnosed patients as "possibly suspicious." In the reader study, 2 readers classified MRIs with a specificity of 93.62% and 91.49%, respectively, and missed 0 and 1 cancer, respectively. On the other hand, the DL model classified MRIs with a specificity of 19.15% and missed 0 cancers, highlighting its potential use not as an independent reader but as a triage tool.

CONCLUSIONS

Our automated DL model triages a subset of screening breast MRIs as "extremely low suspicion" without misclassifying any cancer cases. This tool may be used to reduce workload in standalone mode, to shunt low suspicion cases to designated radiologists or to the end of the workday, or to serve as base model for other downstream AI tools.

Deep learning radiomics for prediction of axillary lymph node metastasis in patients with clinical stage T1-2 breast cancer

Background

This study investigates whether deep learning radiomics of conventional ultrasound images can predict preoperative axillary lymph node (ALN) status in patients with clinical stages T1-2 breast cancer (BC).

Methods

This study retrospectively analyzed the preoperative ultrasound data of 892 patients with BC, who were classified into training (n=535), validation (n=178), and test (n=179) cohorts. Linear combinations of the selected features were weighted by their coefficients to obtain the predicted score. Then, deep learning radiomic features were extracted from the ultrasound images to evaluate the ALN status. Receiver-operating characteristic curves were drawn, followed by the calculation of the area under the curve (AUC) to assess the accuracy of the prediction model in predicting axillary lymph node metastasis (ALNM) in the three cohorts.

Results

Deep learning radiomics combined with radiomics and clinical parameters was the optimal diagnostic predictor of the ALN status in the absence and presence of ALNM, with the AUC of 0.920 (95% confidence interval: 0.872 and 0.968, respectively). Additionally, this combination could also differentiate low-load ALNM [N + (1-2)] from heavy-load ALNM with ≥3 positive nodes [N + (≥3)] in the test cohort, with the AUC of 0.819 (95% confidence interval: 0.568 and 1.00, respectively).

Conclusions

Conclusively, deep learning radiomics of ultrasound images is a non-invasive approach to predicting preoperative ALNM in BC.

Identification of ocular refraction based on deep learning algorithm as a novel retinoscopy method

BACKGROUND

The evaluation of refraction is indispensable in ophthalmic clinics, generally requiring a refractor or retinoscopy under cycloplegia. Retinal fundus photographs (RFPs) supply a wealth of information related to the human eye and might provide a promising approach that is more convenient and objective. Here, we aimed to develop and validate a fusion model-based deep learning system (FMDLS) to identify ocular refraction via RFPs and compare with the cycloplegic refraction. In this population-based comparative study, we retrospectively collected 11,973 RFPs from May 1, 2020 to November 20, 2021. The performance of the regression models for sphere and cylinder was evaluated using mean absolute error (MAE). The accuracy, sensitivity, specificity, area under the receiver operating characteristic curve, and F1-score were used to evaluate the classification model of the cylinder axis.

RESULTS

Overall, 7873 RFPs were retained for analysis. For sphere and cylinder, the MAE values between the FMDLS and cycloplegic refraction were 0.50 D and 0.31 D, representing an increase of 29.41% and 26.67%, respectively, when compared with the single models. The correlation coefficients (r) were 0.949 and 0.807, respectively. For axis analysis, the accuracy, specificity, sensitivity, and area under the curve value of the classification model were 0.89, 0.941, 0.882, and 0.814, respectively, and the F1-score was 0.88.

CONCLUSIONS

The FMDLS successfully identified the ocular refraction in sphere, cylinder, and axis, and showed good agreement with the cycloplegic refraction. The RFPs can provide not only comprehensive fundus information but also the refractive state of the eye, highlighting their potential clinical value.