Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

A collection input based support tensor machine for lesion malignancy classification in digital breast tomosynthesis.

Benjuan Yang1,2, Yingjiang Wu3, Zhiguo Zhou4,2,5

  • 1School of Mathematics and Sciences, Guizhou Normal University, Guiyang 50001, People's Republic of China.

Physics in Medicine and Biology
|November 8, 2019
PubMed
Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

CT-number-preserving bidirectional cycle GAN for contrast addition and removal in head-and-neck radiotherapy planning CT: A multicenter retrospective cohort study.

Radiotherapy and oncology : journal of the European Society for Therapeutic Radiology and Oncology·2026
Same author

Acidosis-associated gene signature defines novel subtypes and dual-target therapeutic candidates in breast cancer.

Journal of translational medicine·2026
Same author

[A Transformer-based multimodal model for predicting hospital-acquired infections using imaging and clinical laboratory data].

Nan fang yi ke da xue xue bao = Journal of Southern Medical University·2026
Same author

A comprehensive multi-task deep learning model for kidney cancer: histological subtyping, clinical staging, and anatomical complexity grading.

European radiology·2026
Same author

Dynamic Chest Radiography as an Alternative to Pulmonary Function Tests for Chronic Obstructive Pulmonary Disease.

Radiology·2025
Same author

CTA-Based Carotid Plaque Reporting and Data System Classification Improves Risk Stratification for Ischemic Cerebrovascular Events.

AJNR. American journal of neuroradiology·2025
This summary is machine-generated.

This study introduces a new computational method to help doctors classify breast lesions as benign or malignant using digital breast tomosynthesis images. By accounting for the unique way these 3D images are captured, the model provides more consistent and accurate diagnostic support.

Area of Science:

  • Medical imaging informatics within diagnostic radiology
  • Computational support tensor machine applications in oncology

Background:

No prior work had resolved the diagnostic inconsistency inherent in processing pseudo-3D volumetric imaging data. Digital breast tomosynthesis provides enhanced lesion visibility but relies heavily on subjective human interpretation. That uncertainty drove the need for automated classification tools to standardize clinical assessments. Prior research has shown that tensor-based models effectively analyze raw imaging data for various diagnostic tasks. However, these conventional frameworks often struggle with the non-uniform resolution found in tomosynthesis datasets. Specifically, the slice spacing differs significantly from the in-plane resolution, creating structural imbalances. This gap motivated the development of specialized algorithms that respect the unique geometry of these scans. Researchers sought to overcome these limitations by proposing a novel approach that avoids forcing inconsistent dimensional structures.

Purpose Of The Study:

The study aims to develop an automatic lesion malignancy classification model to improve diagnostic consistency among physicians using digital breast tomosynthesis. Current diagnostic processes rely heavily on individual physician experience, which can lead to variability in clinical outcomes. The researchers sought to create a computational framework that processes original imaging data directly. They identified that conventional tensor-based classifiers struggle with the pseudo-3D nature of tomosynthesis scans. Specifically, the coarse slice spacing creates dimensional inconsistencies when using standard third-order tensor constructions. This motivation drove the team to introduce a collection input based support tensor machine. The authors intended to relate the third-dimension structural relationship through dynamic weight parameters instead of direct geometric integration. This design ensures that the model remains consistent with the physical properties of the imaging data. The project ultimately strives to provide a more effective tool for identifying malignant breast lesions.

Keywords:
breast cancer screeningvolumetric imaging analysisdiagnostic classification modelmachine learning radiology

Frequently Asked Questions

The researchers propose a collection input based support tensor machine that relates the third-dimension structural relationship via weight parameters in the decision function. This approach avoids the dimensional inconsistency found in conventional classifiers when processing pseudo-3D volumetric data.

The authors utilize a collection input based support tensor machine, which they compare against a kernelled support tensor machine. The former dynamically constructs weights during training, whereas the latter fails to account for the non-uniform resolution property of digital breast tomosynthesis.

The authors state that the third-dimension structural relationship is necessary to relate via weight parameters because the slice spacing is much coarser than the in-plane resolution, which would otherwise lead to inconsistency across all three dimensions.

Related Experiment Videos

Main Methods:

The researchers designed a collection input based support tensor machine to process volumetric imaging data. They utilized a dataset containing 926 total images for training and validation purposes. This collection included 262 malignant cases and 664 benign instances to ensure a diverse sample. The team implemented a decision function that dynamically constructs weight parameters during the training phase. This approach relates the third-dimension structural relationship without forcing a direct geometric construction. They performed a comparative analysis against the kernelled support tensor machine to evaluate performance gains. The study focused on maintaining consistency across all three dimensions of the pseudo-3D data. This methodology prioritizes the unique non-uniform resolution properties inherent in the imaging modality.

Main Results:

The collection input based support tensor machine demonstrated superior effectiveness in classifying breast lesion malignancy compared to the kernelled support tensor machine. Experimental evaluations utilized a comprehensive dataset of 926 images to validate the proposed architecture. The study successfully categorized 262 malignant lesions and 664 benign lesions within the test set. The results indicate that the new model outperforms the latest existing tensor-based classification techniques. By dynamically constructing weight parameters, the system achieved higher consistency across the pseudo-3D volumetric data. The analysis confirms that the method effectively addresses the challenges posed by coarse slice spacing. The findings show that the proposed classifier provides a reliable tool for automated diagnostic support. This performance improvement highlights the utility of the collection input approach for medical image analysis.

Conclusions:

The authors propose that their collection input based support tensor machine effectively addresses the non-uniform resolution of tomosynthesis data. This model provides a more consistent framework for classifying lesion malignancy compared to standard approaches. The researchers demonstrate that their method outperforms existing kernelled support tensor machine architectures in diagnostic accuracy. These findings suggest that incorporating structural relationships through dynamic weight parameters improves classification performance. The study confirms that the proposed technique successfully handles the pseudo-3D nature of breast imaging scans. The authors indicate that their approach offers a robust alternative for automated diagnostic support systems. The results provide evidence that the method is suitable for distinguishing between malignant and benign breast lesions. This work highlights the importance of matching computational models to the specific physical properties of medical imaging data.

The study uses a dataset of 926 images, consisting of 262 malignant and 664 benign cases. This data serves as the input for training the classifier and evaluating its performance against existing tensor-based methods.

The researchers measure the effectiveness of their model by comparing its classification performance against the kernelled support tensor machine. The results illustrate that their method achieves superior outcomes in distinguishing between malignant and benign breast lesions.

The authors propose that their model is effective for classifying breast lesion malignancy in digital breast tomosynthesis. They conclude that this approach outperforms previous methods by better aligning with the pseudo-3D nature of the imaging data.