title: Flow Morphometry of Red Blood Cell Storage Quality Based on Neural Networks
creator: Böcker, Clemens
subject: ddc-600
subject: 600 Technology (Applied sciences)
description: Red blood cell transfusion is routinely performed to improve tissue oxygenation in patients with decreased hemoglobin levels and oxygen-carrying capacity. Generally, blood banks process and store packed red blood cells as RCCs. During storage, RBCs undergo progressive biochemical and morphological changes which are collectively described as storage lesion. According to regulatory guidelines, the quality of RCCs is assessed by quantifying hemolysis before transfusion. However, the hemolysis level only gives an indication of the already lysed erythrocytes; it does not indicate the degree of deterioration of aged cells, which are known to compromise the post-transfusion survival. Morphological analysis, a method that has the potential to provide a simple and practical diagnosis, is suitable for indicating the degradation of RBCs and thus has considerable power to predict actual post-transfusion survival.  Microfluidic systems with suspended RBCs can enable fully automated morphological diagnosis based on image analysis with large cell statistics and high sample throughput. The previous version of the flow morphometry system, which was based on a binary decision tree was able to show in a first attempt that spherocytes are a suitable candidate for such a morphological storage lesion marker. However, due to the low classification resolution (three morphology classes), possible shear-induced morphology changes of the measurement system could not be evaluated.  In this study, the image classification of the flow morphometry system was substantially enhanced by using a convolutional neural network to strongly improve the resolution and accuracy of the morphology classification. The resulting CNN-based classification achieved a high overall accuracy of 92% with RBCs being classified into nine morphology classes. Through this improved classification resolution, it was possible to assess degradation-induced morphologies at high resolution simultaneously with shear-induced morphologies in RCCs. The overall goal was to provide a robust and strong marker for storage lesion that reflects post-transfusion survival of RBCs. Therefore, it was necessary to analyze the extent to which the shear in the microfluidic system affected the morphological transients between RBC classes.  Indeed, it could be shown that shear-induced morphology changes appear dependent on the position of the focal plane height in the flow chamber. The proportion of stomatocytes is increased near the surfaces of the laminar flow chamber. This temporary shear-induced morphology transformation can occur only in flexible erythrocytes with intact membrane properties. Therefore, these cells should be considered a subset of healthy erythrocytes that can reversibly alter from stomatocyte to discocyte morphology.  The nine RBC morphology classes of the improved classification resolution were further analyzed to determine whether they exhibit a particular pattern based on their relative proportions during storage that could be used as a storage lesion marker. All individual RBC classes, except for the spherical morphologies, undergo reversible transitions among themselves that are related to the SDE sequence and result in a low signal-to-noise ratio. The proportions of the irreversible spherical morphologies, spheroechinocytes and spherocytes, were defined as the lesion index. This lesion index showed a strong correlation to hemolysis levels. In fact, the correlation between the hemolysis level and the lesion index was so good that it persisted at an individual RCC level. A preliminary lesion index threshold of 11.1% could be established, which is equivalent to a hemolysis threshold of 0.8% established in regulatory guidelines, to assess whether an RCC is of appropriate quality for transfusion. However, the lesion index, besides predicting the hemolysis level, can also be used to generate more information about post-transfusion survival, since it consists exclusively of the RBC morphologies that are removed by the body in a very short time after transfusion in the recipient.  Finally, we translated the newly established lesion index and standard biochemical parameters into a quality assessment of RCC shipped and transported repeatedly on air rescue missions to assess an eventual deterioration of the RBCs. We showed that the quality of RCCs was not inferior to control samples after repeated air rescue missions during storage. German regulations allow RCCs to be stored for 42 days in a temperature range of +2°C to +6°C. Compliance with this regulation can be secured during air rescue missions by means of suitable logistics based on a rotation system. By using efficient cooling devices, the logistics and maintenance of the thermal conditions are both safe and feasible. A well-defined rotation system for the use of RCCs during routine air rescue missions offers a resource-saving option and enables the provision of RCCs in compliance with German transfusion guidelines. This innovative concept enables life-saving prehospital transfusions directly at the incident scene.  CNN-based flow morphometry and the calculated lesion index allow a reliable assessment of RCC quality. The method also decreases the demand for complex laboratory procedures. Therefore, it is highly advisable to include the lesion index as an additional marker for storage lesion in routine clinical practice. Unlike hemolysis, the lesion index may serve as a good indicator of post-transfusion survival. Thus, both measurements together could provide increased safety and efficacy of stored RCCs.
date: 2023
type: Dissertation
type: info:eu-repo/semantics/doctoralThesis
type: NonPeerReviewed
format: application/pdf
identifier: https://archiv.ub.uni-heidelberg.de/volltextserverhttps://archiv.ub.uni-heidelberg.de/volltextserver/33239/1/Dissertation_Clemens_Boecker.pdf
identifier: DOI:10.11588/heidok.00033239
identifier: urn:nbn:de:bsz:16-heidok-332390
identifier:   Böcker, Clemens  (2023) Flow Morphometry of Red Blood Cell Storage Quality Based on Neural Networks.  [Dissertation]     
relation: https://archiv.ub.uni-heidelberg.de/volltextserver/33239/
rights: info:eu-repo/semantics/openAccess
rights: http://archiv.ub.uni-heidelberg.de/volltextserver/help/license_urhg.html
language: eng