Breast Ultrasound Image Reviewed With Assistance of Deep Learning Algorithms

NCT03706534 | Unknown DMC

• 1 other identifier: 300.08-2018-Samsungmedison-S
• Study Type: interventional
• Target: 300
• Locations: 1 country
• Sites: 1

Brief Summary

This study evaluates a second review of ultrasound images of breast lesions using an interactive "deep learning" (or artificial intelligence) program developed by Samsung Medical Imaging, to see if this artificial intelligence will help the Radiologist make more accurate diagnoses.

Conditions

Breast Cancer; Breast Lesions; Breast Mass

Keywords

Breast cancer; Breast Imaging

Outcome Measures

Primary Outcomes (1)

• Concordance rate

  Breast Imaging Reporting and Data System descriptors suggested by S-Detect for Breast are in good agreement with those selected by experts. In other words, the Breast Imaging Reporting and Data System Lexicon values generated by S-Detect for Breast are not statistically different from the consensus of experts. Breast Imaging Reporting and Data System Assessment Category Score: The user makes the final decision on the Assessment Category Score. Using this Score, S-Detect displays the assessment description. Category 0: Incomplete - Need Additional Imaging Evaluation Category 1: Negative Category 2: Benign Category 3: Probably Benign Category 4a: Low suspicion for malignancy Category 4b: Moderate suspicion for malignancy Category 4c: High suspicion for Malignancy Category 5: Highly Suggestive of Malignancy Category 6: Known Biopsy-Proven Malignancy

  2 days

Secondary Outcomes (6)

• Reporting time

  2 day

• Consensus

  2 day

• Accuracy

  7 day

• Sensitivity

  7 day

• Specificity

  7 day

Study Arms (3)

Manual review

ACTIVE COMPARATOR

The images will be reviewed by the radiologists using BIRADS scheme without any assistance of artificial assistance. This review will be done off-line using a separate program in entirely manual mode. During this review, BIRADS descriptor choices by each radiologist and the time it takes for the radiologist to make such decision will be stored. Radiologists also make assessment decision without any intervention from artificial intelligence. 10 radiologists review manually.

Device: Ultrasound Image review with CADe; Device: Ultrasound Image review with CADx; Device: Ultrasound Image manual review; Procedure: Biopsy

Review by S-Detect for Breast

EXPERIMENTAL

The same images will be separately processed by the artificial intelligence system (S-Detect for Breast) by Samsung. The two results, one by the radiologists and the other by artificial intelligence system, will be compared to statistically quantify equivalence (CADe).

Device: Ultrasound Image review with CADe; Device: Ultrasound Image manual review

Review with assistance of S-Detect for Breast

EXPERIMENTAL

Second, the images will be reviewed by the radiologists with the help of artificial intelligence system, which is an interactive tool automatically providing recommendations on BIRADS descriptor choices that can be modified by the radiologists. The radiologists, after selecting all the descriptors of BIRADS, will decide the assessment categories. These decisions will be compared with the ground truths generated from the biopsy results or a 24-month follow-up (CADx).

Device: Ultrasound Image review with CADx; Device: Ultrasound Image manual review; Procedure: Biopsy

Interventions

Ultrasound Image review with CADe DEVICE

This software is a computer-aided detection (CADe) software application, designed to assist radiologist to analyze breast ultrasound images. S-Detect automatically segments and classifies shape, orientation, margin, lesion boundary, echo pattern, and posterior feature characteristics of user-selected region of interest. The device uses deep learning methods to perform tissue segmentation and classification of images.

Also known as: S-Detect, S-Detect for Breast, CADe, Computer-Assisted Detection Device

Manual review; Review by S-Detect for Breast

Ultrasound Image review with CADx DEVICE

This software is also a computer-assisted diagnostic(CADx) software application, designed to assist a medical doctor in determining diagnosis by presenting whether a lesion is malignant in a breast ultrasound image obtained from an ultrasound imaging device.

Also known as: S-Detect, S-Detect for Breast, CADx, Computer-Assisted Diagnostic Device

Manual review; Review with assistance of S-Detect for Breast

Ultrasound Image manual review DEVICE

The images will be reviewed by the radiologists using BIRADS scheme without any assistance of artificial assistance. This review will be done off-line using a separate program in entirely manual mode. During this review, BIRADS descriptor choices by each radiologist and the time it takes for the radiologist to make such decision will be stored.

Also known as: Convetional Ultrasound image

Manual review; Review by S-Detect for Breast; Review with assistance of S-Detect for Breast

Biopsy PROCEDURE

Suspicious lesions found on breast ultrasound are then followed either by ultrasound guided biopsy or ultrasound imaging every 6 months for two years. For those who undergo biopsy, ultrasound provides images which are used to localize the lesion and guide the placement of the biopsy needle. The sample is sent to pathology for diagnosis, while the ultrasound guidance images are stored. For those who have imaging follow-up, ultrasound images of the breast mass are obtained, digitally stored and interpreted by the radiologist typically using BIRADS scheme.

Manual review; Review with assistance of S-Detect for Breast

Eligibility Criteria

• Age: 19 Years+
• Sex: all
• Healthy Volunteers: Yes
• Age Groups: Adult (18-64), Older Adult (65+)

You may qualify if:

• Adult females or males recommended for ultrasound-guided breast lesion biopsy or ultrasound follow-up with at least one suspicious lesion
• Age \> 18 years
• Able to provide informed consent

You may not qualify if:

• Unable to read and understand English
• Unable or unwilling to provide informed consent
• A patient with current or previous diagnosis of breast cancer in the same quadrant
• Unable or unwilling to undergo study procedures
• Subject Characteristics
• Number of Subjects: 300 subjects from 300 separate breast lesions can be acquired. If a subject has more than 1 suspicious lesion, each may be chosen by the radiologist attending as suitable for "second review".
• Vulnerable Subjects: It is unlikely that any UR students or employees will be enrolled unless their primary physician refers them to UR Medicine Breast Imaging at Red Creek for breast ultrasound and a suspicious lesion is found. We do not expect any of these referrals to be from staffs who work directly with the PIs.

Sponsors & Collaborators

• Samsung Medison: lead
• University of Rochester: collaborator

Study Sites (1)

University of Rochester

Rochester, New York, 14642, United States

Location

MeSH Terms

Conditions

Breast Neoplasms

Interventions

Biopsy

Condition Hierarchy (Ancestors)

Neoplasms by Site; Neoplasms; Breast Diseases; Skin Diseases; Skin and Connective Tissue Diseases

Intervention Hierarchy (Ancestors)

Cytodiagnosis; Cytological Techniques; Clinical Laboratory Techniques; Diagnostic Techniques and Procedures; Diagnosis; Specimen Handling; Diagnostic Techniques, Surgical; Surgical Procedures, Operative; Investigative Techniques

Study Officials

• Avice O'Connell

  Department of Imaging Sciences, University of Rochester

  PRINCIPAL INVESTIGATOR

• Kevin Parker

  Department of Electrical & Computer Engineering, University of Rochester

  PRINCIPAL INVESTIGATOR

Study Design

• Study Type: interventional
• Phase: not applicable
• Allocation: RANDOMIZED
• Masking: SINGLE
• Who Masked: OUTCOMES ASSESSOR
• Masking Details: The study consisted of 10 readers with varying levels of training and experience providing analysis on a randomized set of 300 patients' breast ultrasound data with and without S-Detect for Breast. Two reading periods separated by at least 3-week washout, totaling 600 cases analyzed per reader. PI and her associate have knowledge about patients diagnosis and other information. So, they are exclueded in readers for "reviewing". And all breast US images are de-indentified.
• Purpose: DEVICE FEASIBILITY
• Intervention Model: CROSSOVER
• Sponsor Type: INDUSTRY
• Responsible Party: SPONSOR

Model Details: This clinical study performed by multiple reader multiple case (MRMC) study design, where as set of clinical readers evaulate under multiple reading condition. All Interpreting physician(reader) independently read all of the cases. (fully-crossed design).

Study Record Dates

• First Submitted: October 11, 2018
• First Posted: October 16, 2018
• Study Start: September 20, 2018
• Primary Completion: November 30, 2019
• Study Completion: January 31, 2020
• Last Updated: October 29, 2019

Record last verified: 2019-10

Data Sharing

• IPD Sharing: Will not share

Locations

US (1)