Breast Cancer
Conditions
Keywords
Screening, Low- and middle-income countries, Global health, Artificial intelligence, Ultrasound, Point-of-care ultrasound, clinical breast exam
Brief summary
To assess the feasibility, safety, and clinical performance of community-based breast cancer screening incorporating clinical breast exam and short-termed trained examiners performing AI-supported breast POCUS for triage in limited-resource settings.
Detailed description
Following awareness campaign, asymptomatic and symptomatic women are invited to breast cancer screening. The screening intervention consists initially of clinical breast exam (CBE). Those that are positive at CBE or that present with symptoms are further examined with targeted artificial intelligence (AI)-supported breast point-of-care ultrasound (POCUS) on the complaint site. The AI-supported POCUS is performed by clinical nurses or clinical officers that have undergone a short training program and certification process in POCUS examination. In the first stage of the study, the same women will also be examined by an expert on site (breast radiologist). If clinical safety can be determined, the second stage of the study will start in which expert radiologist will not be on site. Ultrasound images will be assessed by breast radiologists that will serve as reference standard (on site in stage 1 and remotely in stage 2). Women positive at POCUS triage (positive ultrasound finding or with predefined alarming clinical symptoms) will be referred to further follow up.
Interventions
DIAGNOSTIC_TESTAI
AI-supported POCUS for triage
Sponsors
Region Skane
Arba Minch University
Study design
Allocation
NA
Intervention model
SINGLE_GROUP
Primary purpose
SCREENING
Masking
NONE
Masking description
The result of AI-supported POCUS performed by nurse/clinical officer will be masked for the expert radiologist in the first stage of the study
Intervention model description
The study will follow Simon's optimal two-stage phase II design (Simon, 1989) with type I error α = 0.05 and power 80% (β = 0.20). The primary endpoint is sensitivity, defined as the proportion of diseased participants correctly classified as positive by non-experts using AI-supported POCUS, with expert breast radiologists as the reference standard. Null hypothesis (H₀): Sensitivity ≤ 0.80, alternative hypothesis (H₁): Sensitivity ≥ 0.90. A sensitivity of 80% is considered the minimum acceptable level for clinical utility, while 90% represents the target performance based on prior pilot data and expert consensus. In stage 1, enrollment continues until 28 diseased cases are confirmed; if ≤23 are correctly classified, the study stops for futility. Otherwise, enrollment proceeds to 97 diseased cases. The null hypothesis will be rejected if ≥83 cases are correctly classified. This design yields a significance level of 0.0475 and power of 80.2%.
Eligibility
Sex/Gender
FEMALE
Age
18 Years to No maximum
Healthy volunteers
Yes
Inclusion criteria
* Asymptomatic women from the age of 30 with no upper age limit * Symptomatic women 18 years or older.
Exclusion criteria
* Individuals unable to comprehend the study information due to language barriers or cognitive impairments.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Sensitivity | Estimated 2 months (from enrollment until sufficient number of positive participants according to expert assessment) | The primary endpoint is sensitivity, defined as the proportion of diseased women correctly classified as positive by non-experts using AI-supported POCUS. Expert breast radiologists will be used as reference standard |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Specificity | 2 months | Proportion of non-diseased participants correctly classified as negative |
| Positive Predictive Value | 2 months | Proportion of diseased participants of those referred for further workup |
| Negative Predictive Value | 2 months | Proportion of true negatives among all negative results. |
| Cancer Detection Rate | 2 months | Proportion of women with detected cancer among all screened women |
| Recall Rate | 2 months | Proportion of women triaged to further work up |
| False Positive Rate | 2 months | Proportion of participants triaged for work up with no detected cancer among all screened participants. |
| Receiver Operating Characteristic and Area Under the Curve | 2 months | Based on the AI-generated probability score (0-1), using standard ROC analysis |
| Optimal threshold analysis | 2 months | Sensitivity and specificity at the point minimizing distance to (0,1) on the ROC curve. |
| Cancer characterstics | 6 months | Distribution of cancer stage and histological type among detected cancers. |
| Breast cancer in stage I-II | 12 months | Proportion of breast cancer in stage I-II compared to historical standard-of-care data (2021-2025). |
| WHO Global Breast Cancer Initiative key performance indicators | 12 months | Proportion of participants with cancer in relation to the WHO Global Breast Cancer Initiative key performance indicators ( ≥60% of breast cancers diagnosed at stage I-II, ≤60 days from first presentation to diagnosis ≥80% completion of recommended treatment). |
| Usability and Acceptability | 2 months | Reported usability and acceptability of the AI-supported POCUS system: qualitative measure based on questionnaires (standard usability scale 1-5 and proportion of technical issues (y/n)). |
Countries
Ethiopia
Contacts
CONTACTKristina KL Lång, MD PhD
CONTACTAbayneh ATT Tunje Tanga, PhD
PRINCIPAL_INVESTIGATORKristina KL Lån g, MD PhD
Lund University
Outcome results
None listed