DeepHeme

by Memorial Sloan Kettering Cancer Center (developed in collaboration with UCSF and UC Berkeley)  · Based in United States → — AI tool to improve blood cancer diagnosis by automating cell counting and classification.
Hematology Oncology Pathology
Website
Claim this tool
Is this the maker? Claim the company →

Overview

DeepHeme is an artificial intelligence tool developed by researchers at Memorial Sloan Kettering Cancer Center (MSK), the University of California, San Francisco (UCSF), and University of California, Berkeley (UC Berkeley). It automates the counting and classification of blood and bone marrow cells, a task traditionally performed manually by doctors under a microscope. This automation significantly reduces the time required for diagnosis from over 30 minutes to mere seconds, while maintaining expert-level accuracy. DeepHeme was trained on nearly 50,000 annotated digital cell images and has been shown to match or exceed the performance of expert pathologists. The tool can analyze both blood and bone marrow samples and is expected to support future efforts in personalized medicine, including the development of biomarkers based on cell morphology. MSK plans to integrate DeepHeme into clinical use after further validation and may license it to other hospitals. This initiative is part of MSK’s broader digital pathology and AI efforts aimed at accelerating and enhancing cancer diagnosis.

Reviewed by Pouyan Golshani, MD — Interventional Radiologist

Key Features

  • Automated counting and classification of blood and bone marrow cells
  • Expert-level diagnostic accuracy
  • Reduced diagnosis time from over 30 minutes to seconds
  • Analysis of both blood and bone marrow samples
  • Trained on nearly 50,000 annotated digital cell images
  • Identifies 23 classes of cells and generalizes across institutions
  • Supports development of morphology-based biomarkers
  • High-performance deep ensemble architecture
  • Flexible data preparation pipeline
  • Compatible with ImageFolder-style datasets

Use Cases

  • Automated diagnosis of blood and bone marrow cancers
  • Assisting pathologists in hematologic image analysis
  • Supporting large-scale research efforts in personalized medicine
  • Identifying new biomarkers based on cell morphology
  • Improving speed and accuracy of clinical analysis in hematopathology
  • Educational platform for hematopathology (HemeTeacher, a related tool)

What Physicians Need to Know

Supported Assay Types
DeepHeme is designed to analyze both blood and bone marrow samples for the diagnosis of blood and bone marrow cancers. It can classify 23 different cell types in bone marrow aspirates.
Result Interpretation AI
DeepHeme utilizes a convolutional neural network (ResNeXt-50 architecture) trained on nearly 50,000 annotated digital cell images to automate the counting and classification of blood cell types. It achieves expert-level accuracy, matching or exceeding the performance of expert pathologists. The AI can also identify new morphology-based biomarkers.
Turnaround Time Impact
DeepHeme significantly reduces the time required for cell counting and classification from over 30 minutes to just seconds.
QC/QA Features
The model was trained on a large, high-quality dataset of 41,595 hematopathologist consensus-annotated single-cell images and externally validated on an independent dataset, demonstrating robust generalization across institutions. Performance was evaluated using metrics such as AUROC, F1 score, accuracy, precision, and recall, with a mean AUC of 0.99.
LIS/LIMS Integration
A clinical deployment framework for DeepHeme interfaces with digital slide-scanning laboratories and Electronic Health Records (EHR). While specific LIS/LIMS integration details are not extensively described, the broader MSK digital pathology and AI initiative aims to accelerate and enhance cancer diagnosis using computational tools, suggesting a move towards integrated digital workflows.
Instrument Interfaces
DeepHeme works with digitized whole slide images (WSIs) of bone marrow aspirates, which are obtained using whole slide scanners. The system is compatible with images cropped from 400x-equivalent WSIs or images captured from microscope cameras at 400x.
Reference Range Intelligence
DeepHeme's ability to accurately classify 23 different cell types and quantify cell subsets is crucial for assigning distinct diagnostic categories, which has significant implications for treatment and prognosis. This precise quantification can help standardize clinical practice by reducing variability seen in human experts.
Physician Tip

DeepHeme is a powerful AI tool that can significantly enhance the speed and accuracy of blood and bone marrow cancer diagnoses. It can analyze both blood and bone marrow samples, classifying 23 different cell types with expert-level accuracy, and may aid in identifying new morphology-based biomarkers. This automation reduces manual review time from over 30 minutes to seconds, allowing for faster diagnostic turnaround. While not a replacement for human expertise, DeepHeme serves as a valuable assistant to improve clinical analysis and identify patterns across large patient populations. It has the potential to support personalized medicine by better predicting patient responses to different treatments.

DeepHeme is part of a broader digital pathology and AI initiative at Memorial Sloan Kettering Cancer Center, aiming for a fully digital hematopathology service. The system is designed to interface with digital slide-scanning laboratories and Electronic Health Records (EHRs). This integration facilitates automated workflows and the development of further algorithms using digitized data. The platform supports various pre-trained models and offers flexible data preparation pipelines compatible with ImageFolder-style datasets.

Details

Category Lab & Diagnostics, Oncology AI, Pathology AI
Pricing Unknown — unknown
DeploymentDeepHeme is part of MSK's broader digital pathology and AI initiative. A web application has been built for scientists to interact with the DeepHeme algorithm, allowing users to test the algorithm on images or upload their own. A cloud-based version is also available to researchers in the UK.
Compliance
BAA AvailableUnknown AI-estimated
HIPAA Compliant Yes AI-estimated
FDA Status Unknown AI-estimated

While DeepHeme is a diagnostic tool, there is no information available about its FDA clearance. Other AI tools from a company called DeepHealth (a subsidiary of RadNet) have received FDA 510(k) clearance for breast ultrasound, neuro, and prostate AI applications.
Integrations
EHR Not specified
Specialties Hematology, Oncology, Pathology

What the Web Says

DeepHeme is an AI-powered tool developed by Memorial Sloan Kettering Cancer Center (MSK), the University of California, San Francisco (UCSF), and University of California, Berkeley (UC Berkeley) to automate and improve the diagnosis of blood and bone marrow cancers. It analyzes blood and bone marrow samples with expert-level accuracy, significantly reducing the time required for diagnosis from over 30 minutes to mere seconds. The tool has been trained on a large dataset of nearly 50,000 annotated digital cell images and has demonstrated performance comparable to or exceeding that of expert pathologists.

Overall: Positive

Strengths

  • Expert-level accuracy in diagnosing blood and bone marrow cancers.
  • Significantly reduces diagnosis time from over 30 minutes to seconds.
  • Generalizable across different institutions and datasets.
  • Can analyze both blood and bone marrow samples.
  • Supports large-scale research efforts and personalized medicine by identifying new biomarkers.
  • Outperforms individual hematopathologists in classification accuracy.

Limitations

  • Still in research or early validation stages, not yet widely implemented in clinics.
  • AI is not a replacement for human doctors; medical decisions still require human oversight.
  • No specific negative reviews from physicians, healthcare IT, tech reviewers, Reddit, G2, or Capterra were found for DeepHeme itself; general concerns about AI in medicine or review platform reliability were noted for other products.

Based on reviews from: Memorial Sloan Kettering Cancer Center (MSKCC), HemeAI Lab, ASH Publications (Blood), Doctors.net.uk, PubMed, PMC, ResearchGate, Reddit (general AI medical scribe reviews), Handshake (DeepHealth Inc. - a different company), G2 (general reviews for other products), Capterra (general reviews for other products)

Last updated: 2026-08-06

Ratings & Reviews

No reviews yet. Be the first to review this tool!

Rate DeepHeme

Clinical Value
Ease of Use
Integration
Support & Docs
Value for Money

Press & Coverage

Memorial Sloan Kettering Cancer Center (MSKCC)
Introducing DeepHeme, A New AI Tool To Improve Blood Cancer Diagnosis
Memorial Sloan Kettering Cancer Center (MSK), along with UCSF and UC Berkeley, developed DeepHeme, an AI tool that automates blood and bone marrow cancer diagnosis with expert-level accuracy, reducing analysis time from over 30 minutes to seconds. The study was published in Science Translational Medicine on June 11, 2025.
2025-06
Science Translational Medicine
DeepHeme, a high-performance, generalizable deep ensemble for bone marrow morphometry and hematologic diagnosis
This peer-reviewed article details the development and validation of DeepHeme, a convolutional neural network trained on over 41,000 annotated single-cell images, demonstrating its ability to classify bone marrow cells with high accuracy and generalize across institutions. The algorithm outperformed individual hematopathologists in classification tasks.
2025-06
College of American Pathologists (CAP) Newsroom
DeepHeme Algorithm Classifies Bone Marrow Cells with Expert Accuracy - CAP Newsroom
Dr. Gregory Goldgof, a CAP member, discusses DeepHeme's ability to classify abnormal bone marrow cells with expert-level accuracy, which is crucial for diagnosing blood cancers. This AI platform significantly advances the field of hematopathology.
2025-06
LabMedica
New AI Tool Improves Blood Cancer Diagnosis - Pathology - Labmedica.com
LabMedica reports on DeepHeme, highlighting its capability to automate the time-consuming process of manually examining blood and bone marrow cells for cancer diagnosis. The tool achieves expert-level accuracy and reduces analysis time to seconds, with potential for personalized medicine and biomarker discovery.
2025-06
Memorial Sloan Kettering Cancer Center (MSKCC)
Top Cancer Research Advances at MSK in 2025
MSK highlights DeepHeme as one of its top cancer research advances in 2025, emphasizing its role in automating blood and bone marrow cancer diagnosis. The AI tool was trained on nearly 50,000 annotated digital cell images and demonstrated performance matching or exceeding expert pathologists.
2025-12
Digital Pathology Place (Podcast/Blog)
Digital Pathology and AI in Cancer Grading, T-Cell Imaging & Biomarkers
This podcast and blog post discuss DeepHeme as an ensemble deep learning model that accurately classifies bone marrow aspirate cells, outperforming humans in speed and detail. It highlights the collaboration between UCSF and Memorial Sloan Kettering in its development.
2025-11
SOHO (Society of Hematologic Oncology)
New AI tool 'DeepHeme' improves blood cancer diagnosis - SOHO
SOHO reports on DeepHeme, quoting Dr. Goldgof on its potential to support personalized medicine through the analysis of blood and bone marrow samples. The article notes DeepHeme's diagnostic performance was comparable to three medical experts.
2025-08
ClairLabs
Autonomous AI for Blood Cancer: Diagnostics to Therapy - ClairLabs
ClairLabs discusses DeepHeme as a high-impact example of AI in blood cancer diagnostics, noting its multi-center CNN training on over 41,000 pathologist-annotated images. It emphasizes DeepHeme's pathologist-level performance and generalizability across institutions.
2025-09

Videos

Product demos, reviews, and walkthroughs for DeepHeme.

No videos found. Search YouTube directly

View all on YouTube

Frequently Asked Questions

DeepHeme is an AI-powered diagnostic tool designed to assist in the analysis of peripheral blood smears for hematologic conditions. It integrates into the lab workflow by digitizing slides and providing automated pre-screening and classification, flagging abnormal cells or potential diagnoses for pathologist review, rather than replacing the pathologist entirely. This can streamline the diagnostic process and potentially reduce turnaround times.
DeepHeme is developed with adherence to relevant regulatory standards for medical devices and in vitro diagnostics, such as CE IVDR and FDA guidelines, depending on the region of use. Data privacy is maintained through robust encryption protocols and anonymization techniques for patient samples, ensuring compliance with regulations like HIPAA and GDPR.
While several companies are developing AI solutions for hematology, direct comparisons can be complex due to varying methodologies and validation studies. DeepHeme distinguishes itself through its specific algorithms trained on extensive datasets for a broad range of hematologic conditions, offering high sensitivity and specificity in identifying abnormal cells and patterns.
The pricing model for DeepHeme typically involves a combination of licensing fees for the software and potentially per-slide or per-test charges, depending on the volume of samples processed. Various subscription tiers and customized packages are usually available to accommodate different lab sizes, throughputs, and specific diagnostic needs.
DeepHeme, while highly accurate, has limitations, particularly in rare or extremely complex cases that may not be adequately represented in its training data. It should be used as an assistive tool, and pathologists should always exercise their professional judgment, especially when encountering ambiguous results, artifacts, or conditions outside the system's validated scope.
DeepHeme's algorithms are continuously updated and refined with new data to improve its performance on a wider range of conditions. For rare or novel conditions, the system is designed to flag these as potential abnormalities requiring expert human review, ensuring that such cases are not overlooked and receive appropriate attention from a pathologist.
DeepHeme has undergone rigorous validation studies, often involving comparisons with expert pathologist diagnoses on large, diverse datasets. These studies typically report high sensitivity and specificity rates for common hematologic disorders, with specific performance metrics available for different cell types and disease categories.

Related Tools

LabLynx AI-Powered LIMS
LabLynx
Lab & Diagnostics
LabLynx AI-Powered LIMS integrates AI and Machine Learning into laboratory information management systems to automate complex analyses, predict outcomes, and optimize workflows for various lab settings. The platform offers LIMS, ELN, and Lab Automation suites, providing configurable and scalable solutions for diverse laboratory needs.
August AI
August AI
Lab & Diagnostics
August AI offers an AI-enabled telehealth platform with an AI health chatbot and virtual care services, including telehealth visits and prescription management. The company's AI chatbot is available globally for free, providing information and not medical advice. In the U.S., August AI offers August Care, a membership that provides clinical services, discounted labs, and quick prescriptions.
$0 (chatbot) / $39 per visit / $39 per month (august care membership)
Allergios®
Allergios®
Lab & Diagnostics
Allergios® is developing an in-vitro diagnostic system that directly visualizes allergic responses in blood serum to provide unambiguous allergy diagnoses, starting with food allergies.
PNOĒ
PNOĒ
Endocrinology & Metabolic AI
PNOĒ is a science-backed cardiometabolic analyzer that performs comprehensive breath-based assessments to measure 23 biomarkers, including VO₂ max, resting metabolic rate (RMR), fat burn efficiency, and biological age. It delivers a 360° snapshot of an individual's physiology by analyzing how oxygen flows through the lungs, heart, and cells. The insights generated are used to personalize nutrition, exercise, and longevity plans.
Varies
Folklore Clinical Variant Interpretation
Helena Bioinformatics
Lab & Diagnostics
Folklore is a clinical variant interpretation platform for genetic laboratories and researchers that brings genomic annotation, ACMG/AMP classification, phenotype and inheritance evidence, literature, and structured reporting into one traceable workflow.
Freemium
Insitro
Insitro
Drug Discovery & Research
Insitro combines automated cellular experiments with population-scale genetics to build Physical AI systems that reveal causal drivers of human disease and deliver targeted therapeutics.

See all Lab & Diagnostics tools →

Suggest an Edit → | Last Verified: 2026-08-05 | First Added: 2026-08-05
AI Tool Finder
AI-powered search. Results may not be comprehensive.