Project Summary/Abstract
Bone marrow aspirates are critical to the diagnosis, staging, and monitoring of hematologic conditions and
cancers (e.g., leukemia, aplastic anemia, sickle cell disease, and metastasis of solid tumors), but 8-50% of
aspirations are unsuccessful due to operator technique, hemodilution, or underlying pathology. Because this
process is manual and error-prone, there is an opportunity to improve patient outcomes by providing real-time
and automated feedback on the sample quality. Cellia Science will enable improved quality and reliability of
bone marrow aspiration procedures by developing a point-of-care screening instrument. Our approach is
based on a recently developed label-free, deep-ultraviolet (UV) technique for cell imaging and analysis.
Preliminary data has shown that the spicules present in a bone marrow aspirate are easily identifiable by their
characteristic deep blue hue in the unstained pseudocolorized UV image—a result of strong light attenuation at
255nm by bone spicules. The resulting deep-UV images can be generated in under 3 minutes, making the
technique suitable for real-time use during aspiration procedures, and are nearly identical to the Giemsa-
stained slides, which take over 45 min to process. Label-free deep-UV imaging can be combined with machine
learning techniques for feature extraction and classification, which will enable automated quality assessment of
aspirate smears without a pathology technician. We will leverage this technology to develop a bone marrow
aspirate quality screening device for use during aspiration procedures. Towards this goal, we proposed in the
parent grant to quantify sensitivity and specificity of spicule detection by deep-UV microscopy and evaluate
concordance of deep-UV aspirate assessment with visual assessment by trained technician, with Giemsa-
stained samples serving as the ground truth. We also proposed the development of a prototype instrument for
automated spicule adequacy assessment to enable adoption of this technique without a specially trained
pathology technician. In this supplement, we propose to evaluate a single-step object detection approach for
spicule identification by creating a customized analysis pipeline based on the YOLOv7 (You Only Look Once)
architecture. We will also perform customer discovery interviews to identify user needs and develop design
input specifications. Finally, we propose to perform further optimization of the optical train and user interface of
the instrument. Successful implementation of this device is expected to increase the fraction of successful
procedures, which will drastically improve the quality of care for the patient. The proposed work will also help
enhance the diversity of the research and entrepreneurial workforce through the support and training of an
early-stage investigator from an underrepresented group.