Development of AI Technology to Support Differential Diagnosis of Hematological Malignancies
Hitachi and Kyushu University Hospital have developed machine learning-based AI technology to support differential diagnosis of hematological malignancies. The technology uses marker positivity rates from flow cytometry (FCM) testing data as features to provide probability-based candidate diseases for 16 classes, including leukemia, lymphoma, and multiple myeloma. Accuracy of AUC 0.9 or higher was confirmed in evaluations using over 500 clinical cases. Future plans involve implementation through joint verification (PoC) with medical institutions.
📋 Article Processing Timeline
- 📰 Published: June 11, 2026 at 22:00
- 🔍 Collected: June 11, 2026 at 13:21
- 🤖 AI Analyzed: June 11, 2026 at 14:55 (1h 34m after Collected)
Hitachi and Kyushu University Hospital have developed machine learning-based AI technology to assist doctors in the differential diagnosis of hematological malignancies, using flow cytometry (FCM) testing.
Since treatments for hematological malignancies vary significantly by disease type, narrowing down candidate diseases appropriately is crucial for treatment selection. While FCM is an essential diagnostic test that measures cellular marker characteristics, interpreting the data requires high expertise and experience, and the workload is increasing due to the rise in caseloads. The developed AI technology uses the 'marker positivity rate' within cell populations as a feature to classify them in a way close to the actual diagnostic process. It supports organization of information and new insights by providing multiple candidate diseases with associated probabilities for 16 classes, including leukemia, lymphoma, and multiple myeloma. Evaluations using over 500 clinical cases at Kyushu University Hospital confirmed performance with an AUC of 0.9 or higher for simultaneous multi-disease classification.
Hitachi aims to scale evaluation through joint verification (PoC) with medical institutions and testing companies, implementing it as a diagnostic support technology that contributes to both quality and sustainability of healthcare.
■ Background and Challenges
With the global increase in cancer, creating mechanisms to support the interpretation of test results by limited medical personnel is becoming important. FCM testing, which measures markers by irradiating cells with laser light, requires high levels of expertise and experience in interpreting results such as 'gating.' Furthermore, due to the increased burden of analysis work accompanying the rise in testing, there was a need for technology to efficiently analyze data and present diagnostic decision-making materials more clearly.
■ Features of the Developed Technology
1. AI model utilizing 'Marker Positivity Rate': Constructed to follow the interpretation process after gating, supporting the standardization of diagnosis based on the indices physicians typically use.
2. Presentation of multiple candidates with probabilities: By providing multiple candidate diseases (currently 16 classes) with probabilities rather than a single diagnosis, it provides decision-making materials, assisting physicians in narrowing down candidates in line with their diagnostic reasoning.
■ Confirmed Effects
Two models covering leukemia, lymphoma, and multiple myeloma were built and evaluated using over 500 clinical cases at Kyushu University Hospital, confirming an AUC of 0.9 or higher.
■ Future Outlook
Hitachi will expand evaluation through PoCs with medical institutions, aiming for smooth utilization in clinical workflows. Part of these results were published as an abstract at the European Hematology Association (EHA) 2026 Congress held in Sweden from June 11 to 14.
Since treatments for hematological malignancies vary significantly by disease type, narrowing down candidate diseases appropriately is crucial for treatment selection. While FCM is an essential diagnostic test that measures cellular marker characteristics, interpreting the data requires high expertise and experience, and the workload is increasing due to the rise in caseloads. The developed AI technology uses the 'marker positivity rate' within cell populations as a feature to classify them in a way close to the actual diagnostic process. It supports organization of information and new insights by providing multiple candidate diseases with associated probabilities for 16 classes, including leukemia, lymphoma, and multiple myeloma. Evaluations using over 500 clinical cases at Kyushu University Hospital confirmed performance with an AUC of 0.9 or higher for simultaneous multi-disease classification.
Hitachi aims to scale evaluation through joint verification (PoC) with medical institutions and testing companies, implementing it as a diagnostic support technology that contributes to both quality and sustainability of healthcare.
■ Background and Challenges
With the global increase in cancer, creating mechanisms to support the interpretation of test results by limited medical personnel is becoming important. FCM testing, which measures markers by irradiating cells with laser light, requires high levels of expertise and experience in interpreting results such as 'gating.' Furthermore, due to the increased burden of analysis work accompanying the rise in testing, there was a need for technology to efficiently analyze data and present diagnostic decision-making materials more clearly.
■ Features of the Developed Technology
1. AI model utilizing 'Marker Positivity Rate': Constructed to follow the interpretation process after gating, supporting the standardization of diagnosis based on the indices physicians typically use.
2. Presentation of multiple candidates with probabilities: By providing multiple candidate diseases (currently 16 classes) with probabilities rather than a single diagnosis, it provides decision-making materials, assisting physicians in narrowing down candidates in line with their diagnostic reasoning.
■ Confirmed Effects
Two models covering leukemia, lymphoma, and multiple myeloma were built and evaluated using over 500 clinical cases at Kyushu University Hospital, confirming an AUC of 0.9 or higher.
■ Future Outlook
Hitachi will expand evaluation through PoCs with medical institutions, aiming for smooth utilization in clinical workflows. Part of these results were published as an abstract at the European Hematology Association (EHA) 2026 Congress held in Sweden from June 11 to 14.
FAQ
What is the accuracy of Hitachi's AI technology for diagnosing 16 classes of hematological malignancies?
Hitachi's AI technology achieved an accuracy of AUC 0.9 or higher in diagnosing 16 classes of hematological malignancies.
How many clinical cases were used to evaluate the performance of Kyushu University Hospital's AI diagnostic tool?
Over 500 clinical cases were used to evaluate the performance of Kyushu University Hospital's AI diagnostic tool.
Which medical institution collaborated with Hitachi on the development of the AI-based differential diagnosis system in 2023?
Kyushu University Hospital collaborated with Hitachi on the development of the AI-based differential diagnosis system in 2023.
What type of data does Hitachi's AI technology use from flow cytometry testing for leukemia diagnosis?
Hitachi's AI technology uses marker positivity rates from flow cytometry testing data for leukemia diagnosis.
What is the planned next step for implementing the AI technology developed by Hitachi and Kyushu University Hospital?
The planned next step is implementation through joint verification (PoC) with medical institutions.