What does this research mean for the field?

Low-dose X-ray diffraction imaging with real-time AI classification can distinguish ductal carcinoma in situ (DCIS) from invasive breast cancer with high accuracy. Novelty: ClaimNovelty.NOVEL_FINDING. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To evaluate the performance of EoScan™ in differentiating DCIS from invasive breast cancer using low-dose X-ray diffraction imaging and AI classification.

February 19, 2026

Abstract PS1-06-15: Low-Dose X-Ray Diffraction Imaging with Real-Time AI Classification Distinguishes DCIS from Invasive Breast Cancer: Hardware and Interim Clinical Performance from a First-in-Human Study

Key Points

To evaluate the performance of EoScan™ in differentiating DCIS from invasive breast cancer using low-dose X-ray diffraction imaging and AI classification.
Ex-vivo cohort study with 161 malignant and 131 benign breast samples collected.
Utilized a convolutional neural network trained on over 12,700 diffraction patterns.
In-vivo testing with 150 participants using the EoScan™ device to obtain scans and assess performance metrics.
Ex-vivo sensitivity of 95.9% and specificity of 93.5%, with an accuracy of 90% for distinguishing DCIS from invasive cancer.
Interim in-vivo sensitivity of 85% and specificity of 70% across 150 participants.
High patient satisfaction reported with no device-related adverse events.

Abstract

Abstract Low Dose X-Ray Diffraction Imaging with Real Time AI Classification Distinguishes DCIS from Invasive Breast Cancer: Hardware and Interim Clinical Performance of the EosDx EoScan™ First-in-Human StudyAika Tanaka1, Audrey Nguyen1, Mark Pegram2, Masroor Khonkhodzhaev3,4, Slava Shcherbakov4, DarylHoffman5, Alexander Lazarev1, Byron Aram1, Thomas J. Lomis6, Prashant Chawla6, Lev Mourokh3,Pavel Lazarev11 EosDx, Inc., Menlo Park, CA, USA 2 Stanford Cancer Institute, Stanford University CA, USA 3 PhysicsDepartment, Queens College of the City University of New York, NY, USA, 4 Matur UK Ltd., London,UK, 5 Daryl Hoffman MD, Campbell, CA USA, 6 San Fernando Valley Cancer Foundation, Van Nuys, CA,USA Background Xray diffraction detects nanoscale structural alterations in collagen, lipid, and water that heraldbreast neoplasia. EoScan™ is a diffraction breast scanner that captures Bragg signatures throughintact tissue and classifies malignancy using a convolutional neural network (CNN) trained on12,700 spectra. First-in-Human Hardware The IRB-approved first-in-human unit ("Human-1", IRB #1381803, San Fernando Valley CancerCenter, Los Angeles, CA) comprises: (1) a micro-focus Xray source with liquid cooling, (2) a80 µm slot collimator, (3) a 256×256-pixel photon-counting detector housed in a 2 mmPb-equivalent shield. The system fits in 0.75 m2, weighs 110 kg, runs standard 110V, withoutbreast compression. Methods Ex-vivo cohort: Breast samples including malignant (n = 161) and benign (n = 131) obtained atValley Breast Care (Los Angeles, USA) and Keele University (Keele, UK). A total of 12,730diffraction patterns were collected at EosDx (Menlo Park, CA) and Queen Mary (London, UK)respectively. Malignant samples include infiltrating and in-situ types. Healthy control tissue wasdonated by patients undergoing reconstructive surgery performed by Dr. Daryl Hoffman(Campbell, CA); all samples were collected under the same IRB-approved protocol governing thein-vivo study. For all samples, pathology testing was conducted by the Valley Breast CarePathology Department to confirm the disease status. Standard immunohistochemistry testing isconducted for information on hormone receptor status. FISH results are also reported. In-vivo cohort: ongoing NOVA first-in-human trial (n = 150) using the Human-1 device atSan Fernando Valley Cancer Center; patients are guided into exam room after completing IRBapproved consent form and receive 3 scans on each breast. CNN outputs a malignancy index (0–100). Radiologic findings and pathology results are used to validate the malignancy index outputand calculate performance metrics, such as Sensitivity (Se), Specificity (Sp), PPV, and ROC-AUC. DCIS versus invasive carcinoma performance is evaluated separately. Results Ex-vivo: Se = 95.9 %, Sp = 93.5 %, PPV = 95.1 %, AUC = 0.96; DCIS vs invasiveaccuracy = 90 % In-vivo interim: Se = 85 %, Sp = 70 % across 150 participants; collected results are analyzedthrough previously trained machine learning models and cloud-based weekly model updatesimprove performance; no device-related adverse events. Patients report high satisfaction and apositive user experience. Conclusions EoScan delivers molecular level breast imaging at radiation doses ∼25-fold belowmammography. High ex-vivo accuracy and encouraging interim in-vivo performance, along withthe ability to distinguish DCIS from invasive cancer, support its potential as a front line screeningadjunct. Final NOVA results will be presented upon database lock. Keywords X-ray diffraction, breast cancer, DCIS, AI diagnostics, structural biomarkers, low-dose imaging Funding Sponsored by EosDx Inc. Citation Format: A. Tanaka, A. Nguyen, M. D. Pegram, M. Khonkhodzhaev, S. Shcherbakov, D. Hoffman, A. Lazarev, B. Aram, T. J. Lomis, P. Chawla, L. Mourokh, P. Lazarev. Low-Dose X-Ray Diffraction Imaging with Real-Time AI Classification Distinguishes DCIS from Invasive Breast Cancer: Hardware and Interim Clinical Performance from a First-in-Human Study abstract. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS1-06-15.

Bookmark

Abstract PS1-06-15: Low-Dose X-Ray Diffraction Imaging with Real-Time AI Classification Distinguishes DCIS from Invasive Breast Cancer: Hardware and Interim Clinical Performance from a First-in-Human Study

Key Points

Abstract

Cite This Study