Leonie Saft, PhD, is a Senior Consultant Hematopathologist at the Department of Clinical Pathology and Cancer Diagnostics, Karolinska University Hospital in Stockholm. She is team leader of the division of hematopathology and flow cytometry laboratory.
Dr. Saft and her team wanted to create an automated tool that could replace the manual differential count in bone marrow smears. We interviewed Dr. Saft about the project and her journey with Aiforia so far. Continue reading to learn more or skip to the video interview at the end of this article.
“Machine-learning based algorithms in digital pathology are particularly useful for the automatization of certain diagnostic steps like the quantification of hematopoietic cells in bone marrow aspirate smears.” – Leonie Saft, Senior Consultant, Karolinska University Hospital
Background of the project
The assessment of bone marrow aspirate smears plays an important role in the routine diagnostic workup of patients with hematological disease, and is usually combined with the histological examination of bone marrow biopsies. The cytomorphological assessment is often the first step in the acute clinical setting of unclear cytopenia and suspicious leukemia.
Bone marrow smears or imprints are the basis for performing manual differential cell counts of different cell types within the main hematopoietic lineages. They are particularly important for the quantification of immature cells, so called blasts, but also for the qualitative assessment of hematopoietic cells.
The manual differential count of 500 cells, as recommended by the WHO and the International Council for Standardization in Hematology (ICSH), is labor-intensive, time-consuming, and requires high expertise. “This is why we aimed at developing a model for automated cell detection, similar to cell counters that are in use for leukocyte differential counts in the peripheral blood. However, those are not applicable to the bone marrow,” Dr. Saft explains.
Developing the AI model
The team first selected 20 archived bone marrow aspirate smears from patients with normal or reactive findings in the bone marrow. The digitized slides were pseudonymized and uploaded to the Aiforia® Platform. Aiforia’s AI development tool, Aiforia® Create, was used to train the model. It was designed to detect nine major cell types.
Single-cell annotations were performed in training regions to define the “ground truth” and repeated until the model achieved adequate performance. External human validation was performed independently by three experts in bone marrow cytology.
The cloud environment of the Aiforia® Platform allowed the team to work remotely without the need for software installation. As Dr. Saft describes it:
“The biggest benefit of using the Aiforia® Platform is that it's cloud-based and that it did not require installation or any sophisticated equipment at our laboratory. The microscopic slides that were used for this project were digitized and uploaded to the platform. The only requirement was that the slides were stained using the same staining protocol and that the same scanner was used. Another advantage was that the training and external validation was completely remote and could be performed at any place and at any time, and this was clearly a big advantage.”
Experience working with Aiforia
“My experience with Aiforia is very positive. The Aiforia® Platform is user-friendly and has inbuilt support with the quick start guide to Aiforia and also a Knowledge Base with a lot of information on how to create an AI model. I started my project a couple of years ago and had continued support from two Aiforia experts throughout the project either by mail or through scheduled Zoom meetings.”
As Dr. Saft is working full-time clinically and did not have any dedicated time period reserved for the project, it was extremely important that support was always available. “I relied very much on quick and immediate support, which I always received,” she explains.
As for what is next, Dr. Saft would like to continue working with Aiforia on other projects within clinical hematopathology.
Watch Dr. Saft’s interview from the video:
Back to all blog posts
