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

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
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
HIPAA Compliant Yes AI-estimated
FDA Status Unknown
Integrations
EHR Not specified
Specialties Hematology, Oncology, Pathology

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
No press coverage or publications on file yet. Know of some? Suggest an edit →

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

PathologyAILab
PathologyAILab
Lab & Diagnostics
PathologyAILab provides AI-enhanced digital pathology analysis and diagnostic insights for healthcare providers, utilizing a multi-agent AI system for faster, more accurate cytology results.
Kneevoice Cartilage Evaluation System (750-3600-001)
Kneevoice
Lab & Diagnostics
Kneevoice develops innovative, non-invasive diagnostic devices to assess knee cartilage health by capturing and analyzing sounds produced by the knee joint during movement.
myosmart.cuff (757M20-2)
Otto Bock Healthcare Products GmbH
Lab & Diagnostics
The myosmart.cuff is a medical device by Otto Bock Healthcare Products GmbH designed for objective measurement of muscle tone, primarily for use in neurological rehabilitation.
Seaman Pro/Seaman
Checkcells
Lab & Diagnostics
CheckCells' Seaman PRO™ is an AI-powered microscope for automated semen analysis, offering high accuracy and efficiency for fertility diagnostics in clinics and at-home testing.
LensHooke X1 Pro Semen Quality Analyzer, LensHooke X1 Semen Quality Analyzer
Bonraybio Co.
Lab & Diagnostics
The LensHooke X1 Pro Semen Quality Analyzer and LensHooke X1 Semen Quality Analyzer are automated, compact devices that use AI optical microscopy for rapid and reliable human semen analysis, providing objective results for various sperm parameters.
Vitestro
Vitestro
General Productivity (AI Assistants)
Aletta by Vitestro is the world's first Autonomous Robotic Phlebotomy Device (ARPD) that automates diagnostic blood collection to improve standardization and address staffing shortages.

See all Lab & Diagnostics tools →

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