GE HealthCare, DeepHealth Expand Mammography Collaboration

tim.hodson

View Original ↗
AI 導讀 industry AI 重要性 3/5

緻密乳房偵測率提升 23%:GE 與 DeepHealth 把 AI 乳攝多階段工作流推向國際市場

  • 579,583 筆真實篩檢驗證:多階段 AI 工作流讓整體乳癌偵測率相對提升 21%,緻密乳房族群更達 23%
  • ProFound Pro 自動病灶定位 + 密度分類、Safeguard Review 自動標記複雜案例進入 second reader,兩者組合覆蓋前線偵測與後端複查
  • ProFound Pro 僅限美國、Safeguard Review 尚未 CE 認證,臺灣機構採購前須向 GE 當地代理確認可用性與基準偵測率

乳攝篩檢的漏診問題常被低估——在美國橫跨 4 州、579,583 筆篩檢的真實世界研究中,多階段 AI 工作流讓整體乳癌偵測率比純人工閱片高出 21%;聚焦緻密乳房族群,差距更擴大至 23%。2026 年 4 月,GE HealthCare 與 DeepHealth(RadNet 旗下全資子公司)宣布深化合作,準備將這套工具推向美國以外的國際市場。

從 2024 到國際化:合作擴張的三大進展

2024 年,GE HealthCare 與 DeepHealth 首度攜手,將 DeepHealth 的 AI 乳癌篩檢流程整合進 GE 的 Senographe Pristina 乳攝系統。目標是讓同一張 FFDM(full-field digital mammography,全視野數位乳攝)在拍攝後立刻進入智慧篩查管線,不需醫院另行搭建複雜的中介系統,降低導入門檻。

2026 年 4 月宣布的新一輪擴張有三個實質進展:其一,合作範圍從美國正式延伸至國際市場;其二,增加 second reader 工作流(第二讀片流程),讓高風險案例能自動進入雙重審視路徑;其三,DeepHealth 新一代旗艦套件 Breast Suite 的兩款核心模組——ProFound Pro 與 Safeguard Review——正式納入 GE 的產品發行體系,在 GE 自有乳攝系統上直接提供。時機選在乳房影像學會(SBI)年度研討會週,是向業界宣示商業布局的明確信號。

ProFound Pro 與 Safeguard Review:乳攝 AI 兩階段防護

理解這次合作的核心,需要先拆開 Breast Suite 的架構。這是一套雲端優先、模組化的應用程式集合,設計上直接對接現有乳攝工作流,無需醫院替換 PACS 或 RIS,以模組化方式滾動導入。

ProFound Pro 是前線診斷模組,整合兩個功能:第一是 Cancer Detection(癌症偵測),自動定位病灶並給出懷疑程度評分(degree of suspicion),對緻密乳房組織及多元族群均有一定效能;第二是 Automated Density Assessment(自動乳腺密度分類,可理解為「讓 AI 代替醫師主觀判斷緻密程度」),對 2D 或 3D 乳攝影像進行一致性 BI-RADS 密度分類,降低判讀者間的個人差異。

Safeguard Review 是第二層保險,設計邏輯截然不同:它不直接診斷,而是自動標記「可能受益於二次審視」的複雜案例,推入 second reader 工作流。這相當於一個自動分診系統——把高風險案例攔出來,防止因工作量大而讓重要病灶悄悄流過。此外,Timely Alerts 整合進 Pristina Via(GE 新一代乳攝系統)後,能在拍片後 5 分鐘內發出高懷疑度案例警示,讓醫師不需等到批量閱片完成才識別需要即時處理的緊急案例。

Breast Suite 核心模組功能與地區供應對照
模組核心功能地區限制
ProFound Pro — Cancer Detection自動病灶定位 + 懷疑程度評分,適用緻密乳房及多元族群僅限美國
ProFound Pro — Automated Density Assessment2D/3D 乳攝自動 BI-RADS 密度分類,降低判讀者主觀差異僅限美國
Safeguard Review自動標記複雜案例進入 second reader 工作流,降低漏診尚未 CE 認證
Timely Alerts(整合 Pristina Via)5 分鐘內警示高懷疑度案例部分市場提供

Breast Suite 為模組化雲端套件,非所有功能均在美國以外市場提供

579,583 筆篩檢:Nature Health 研究的規模與設計

支撐這次合作的核心證據,來自 Louis LD、Wakelin EA、McCabe MP 等人發表於《Nature Health》(2025)的大型真實世界研究。研究全名強調「equitable impact」(公平性影響),代表設計之初即涵蓋多元族群樣本,而非只在單一學術醫學中心驗證。

研究規模:579,583 筆乳攝影像、4 個州、109 個篩檢站點、96 位放射科醫師。多階段 AI 工作流的運作方式是:AI 先篩出高懷疑度案例,再交由乳攝專科醫師(由各機構乳房影像科主任按臨床表現和資歷精選)進行第二輪讀片。對照組(standard of care)定義為放射科醫師在完全無 AI 輔助下的單人讀片。

需要特別注意的是,最終偵測率的提升是「AI 演算法 + 精選 second reader」這個組合的結果,並非純 AI 單獨的貢獻。新聞稿中引用的 21%相對提升率,基準絕對偵測率未在商業宣告層級揭露——臨床評估時,應要求廠商提供原始研究的每千人癌症偵測率(cancers per 1,000 screened),才能換算出對本機構的實際意義。

AI 多階段工作流 vs 傳統閱片:乳癌偵測率相對提升(%)

資料來源:Louis et al., Nature Health 2025(n=579,583,4 州,109 站點,96 位放射科醫師)

緻密乳房的 23%:人工最吃力、AI 最有效的交叉點

子群分析顯示,緻密乳房(dense breast)族群的偵測率相對提升達 23%,高於整體的 21%。從機制看完全符合預期:緻密乳腺組織(BI-RADS C/D 類)在傳統乳攝上產生「白遮白」效應,腫塊與纖維腺體組織密度相近,人眼難以辨別邊界,是傳統讀片遺漏率最高的場景之一。

AI 系統不依賴整體灰階對比做判斷,而是學習大量影像中的局部特徵——邊緣不規則性、毛刺徵(spiculated margins)、微鈣化群集模式——這恰好是傳統讀片最吃力的地方,也是 AI 相對優勢最為突出的情境。換句話說,AI 的偵測增益並非均勻分布,而是集中在人工最容易出錯的子群。

Automated Density Assessment 的臨床意義在此進一步延伸:系統性識別高密度乳房患者後,臨床端可主動規劃補充影像(超音波或乳房 MRI),而不是在下一輪篩檢才偶然發現漏診。這形成一個完整的風險分層迴圈:乳攝 → 密度評估 → 高密度標記 → 補充影像建議,把 AI 從「輔助閱片工具」提升為「風險分層的起點」。

部分模組僅限美國:地區限制與基準值的現實門檻

這次合作宣告中有幾個關鍵細節,臨床部署前必須釐清,不能只看標題數字就進行採購評估。

地區限制明確。 ProFound Pro 目前僅在美國提供;Safeguard Review 尚未通過 CE 認證,在歐洲及絕大多數非美國市場無法使用。臺灣市場的實際供應狀況,需向 GE HealthCare 當地代理商直接確認,不能以美國的商業宣告作為採購依據。

21% 是相對提升,基準值未公開。 相對提升率的臨床意義取決於起始基準。若機構基準偵測率已在業界高水準,21% 相對增幅換算成絕對值可能有限;若偵測率本偏低,空間則相當可觀。採購評估時應要求廠商提供基準絕對偵測率,不能只比較相對百分比。

Second reader 是精選醫師,非隨機抽取。 研究中參與二次讀片的醫師由科主任按臨床表現精選,這樣的設計通常會高估一般機構部署 AI 後的真實效益。Safeguard Review 的實際分診效益,取決於機構能否同步建立一個有品質保障的二次讀片制度——若只有 AI 分診但沒有配套的讀片資源,流程效益會大打折扣。

AI 仍需人工決策閉環。 所有模組的最終判讀仍由放射科醫師負責,AI 的角色是降低漏診機率並優化分流,不能用「有 AI 就可以縮減人力」的邏輯規劃人力配置,這是業界在導入任何 AI 輔助工具時最常出現的誤解。

緻密乳房不是人工讀片的罩門——是 AI 能多拿 23% 偵測率的主場。BI-RADS C/D 的密度分類不只是報告附注,是決定要不要啟動補充影像的第一道觸發點。

Abstract

tim.hodson Mon, 04/20/2026 - 17:35 April 16, 2026 – GE HealthCare has expanded its collaboration with DeepHealth, Inc., a wholly-owned subsidiary of RadNet, Inc., "to further the innovation, commercialization and adoption of advanced AI-powered mammography tools."  Photo: Getty ImagesThe announcement coincides with the Society of Breast Imaging (SBI) Annual Symposium. The expanded collaboration builds on the joint initiative first announced in 2024, which united DeepHealth’s AI-powered breast cancer screening workflow solution with GE HealthCare’s Senographe Pristina mammography system to enhance image interpretation and improve operational efficiency. This next phase of the joint initiative will extend the collaboration’s relationship for breast cancer solutions internationally, expanding the offering with second reader workflow and extending access to components of DeepHealth’s breast cancer solutions.1 With the continued growth in capabilities of DeepHealth’s new Breast Suite2 solution, GE HealthCare continues to bring to market an advanced, cloud-based suite of modular, interoperable applications that integrate seamlessly into existing breast imaging workflows. Paired with GE HealthCare’s Pristina Via, the combined offering enables scalable, high-efficiency breast cancer screening programs. “At GE HealthCare, we’re advancing women’s health through precision care built around the unique needs of women and enhanced by the power of AI,” said Jyoti Gupta, PhD, President and CEO, Women’s Health and X-ray at GE HealthCare. “By integrating Breast Suite AI with our Pristina Via mammography system, we’re helping clinicians detect breast cancer early with greater confidence. These innovations move us closer towards truly personalized prevention and care for women.” In the largest real-world analysis of AI-powered breast cancer screening in the U.S., a multistage AI-driven workflow from DeepHealth’s Breast Suite have shown to enable a 21% increase in breast cancer detection rate when compared to the standard of care3. Additionally, Breast Suite’s Timely Alerts features uniquely integrates with Pristina Via to notify sites of potentially suspicious cases in minutes.4   “Pristina Via and Breast Suite have transformed how we approach breast cancer detection,” said Dr. Meghna Krishnan, Owner, URPrecious Imaging, Arizona. “The cloud-based viewer gives us instant access to images from anywhere, and the AI tools, like CAD for lesion detection and automated breast density assessment, can help us deliver fast and consistent results. It’s not just about speed, it’s about confidence. With Breast Suite, our team can make more accurate decisions without workflow disruptions, which ultimately means better care for our patients.” The existing Breast Suite solution offered by GE HealthCare includes a cloud-first multi-modality Viewer, Cancer Detection, Automated Density Assessment, prioritized Worklist, Timely Alerts, and enhanced Reporting. The expanded collaboration will enable GE HealthCare to distribute new Breast Suite applications designed for compatibility with GE HealthCare’s mammography systems, including: ProFound Pro,5 which combines Cancer Detection, a clinical AI solution that offers automatic lesion localization and degree of suspicion that are effective in diverse populations and dense breast tissue, and Automated Density Assessment for consistent, automated density classification with a patient-centric, accurate density assessment of 2D or 3D mammograms to support objective diagnostic decisions. Safeguard Review6: OptionalAI-powered workflow that flags complex cases that may benefit from a secondary review. A large U.S. study demonstrate d a 21% increase in cancer detection rate and a 23% increase in cancer detection rate for women with dense breasts with a multistage AI-driven workflow when compared to the standard of care.1 “At DeepHealth, we are committed to bringing the next era of AI-powered health informatics to help stage shift disease,” said Niccolo Stefani, MD, Business Leader Population Health & Clinical AI, DeepHealth. “Through our expanded collaboration with GE HealthCare, we are bringing the power of our new Breast Suite to providers around the world to enable early cancer detection, deeper clinical insights, and more confident decisions. Together we are pushing the boundaries of what mammography can deliver, not only for breast cancer screening but for women’s health more broadly.” For more information, please visit https://deephealth.com/ and www.GEHealthCare.com   Certain components only available in the U.S. Breast Suite comprises multiple applications including ProFound Pro, Breast Density, and DeepHealth Viewer. DeepHealth Viewer is manufactured by eRAD and distributed by DeepHealth. Not all components and functionality are available in all markets. Any claims made about Breast Suite may reference claims associated with its individual components. Louis, L.D., Wakelin, E.A., McCabe, M.P. et al. Equitable impact of an AI-driven breast cancer screening workflow in real-world US-wide deployment. Nature Health (2025). This study was conducted across 4 states with 579,583 exams, 109 sites, and 96 radiologists. The multistage AI-driven workflow leverages breast imaging specialists selected by breast imaging practice leadership based on experience and clinical performance record for secondary review. The standard of care is based on radiologists reading cases without any AI. Data on file: Rapid image processing flags highly suspicion cases in under five minutes when integrated with GE HealthCare’s Senographe Pristina system and using 1 GB bandwidth transmission. Not CE marked. Not available in all markets. Only available in the U.S. Monday, April 20, 2026 - 17:35