Knowledge transfer to enhance the performance of deep learning models for automated classification of B cell neoplasms.

Multi-parameter flow cytometry (MFC) is a cornerstone in clinical decision making for leukemia and lymphoma. MFC data analysis requires manual gating of cell populations, which is time-consuming, subjective, and often limited to a two-dimensional space. In recent years, deep learning models have been successfully used to analyze data in high-dimensional space and are highly accurate. However, AI models used for disease classification with MFC data are limited to the panel they were trained on. Thus, a key challenge in deploying AI into routine diagnostics is the robustness and adaptability of such models. This study demonstrates how transfer learning can be applied to boost the performance of models with smaller datasets acquired with different MFC panels. We trained models for four additional datasets by transferring the features learned from our base model. Our workflow increased the model's overall performance and, more prominently, improved the learning rate for small training sizes.

Hematologist-Level Classification of Mature B-Cell Neoplasm Using Deep Learning on Multiparameter Flow Cytometry Data.

The wealth of information captured by multiparameter flow cytometry (MFC) can be analyzed by recent methods of computer vision when represented as a single image file. We therefore transformed MFC raw data into a multicolor 2D image by a self-organizing map and classified this representation using a convolutional neural network. By this means, we built an artificial intelligence that is not only able to distinguish diseased from healthy samples, but it can also differentiate seven subtypes of mature B-cell neoplasm. We trained our model with 18,274 cases including chronic lymphocytic leukemia and its precursor monoclonal B-cell lymphocytosis, marginal zone lymphoma, mantle cell lymphoma, prolymphocytic leukemia, follicular lymphoma, hairy cell leukemia, lymphoplasmacytic lymphoma and achieved a weighted F1 score of 0.94 on a separate test set of 2,348 cases. Furthermore, we estimated the trustworthiness of a classification and could classify 70% of all cases with a confidence of 0.95 and higher. Our performance analyses indicate that particularly for rare subtypes further improvement can be expected when even more samples are available for training. © 2020 The Authors. Cytometry Part A published by Wiley Periodicals LLC. on behalf of International Society for Advancement of Cytometry.

Absolute quantification of regional renal blood flow in swine by dynamic contrast-enhanced magnetic resonance imaging using a blood pool contrast agent.

AIM: To evaluate for the first time in an animal model the possibility of absolute regional quantification of renal medullary and cortical perfusion by dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) using a blood pool contrast agent. MATERIAL AND METHODS: A total of 18 adult female pigs (age, 16-22 weeks; body weight, 45-65 kg; no dietary restrictions) were investigated by DCE-MRI. Absolute renal blood flow (RBF) measured by an ultrasound transit time flow probe around the renal vein was used as the standard of reference. An inflatable stainless cuff placed around the renal artery near its origin from the abdominal aorta was used to reduce RBF to 60%, 40%, and 20% of the baseline flow. The last measurement was performed with the cuff fully reopened. Absolute RBF values during these 4 perfusion states were compared with the results of DCE-MRI performed on a 1.5-T scanner with an 8-channel phased-array surface coil. All scans were acquired in breath-hold technique in the coronal plane using a field of view of 460 mm.Each dynamic scan commenced with a set of five 3D T1-weighted gradient echo sequences with different flip angles (alpha = 2 degrees, 5 degrees, 10 degrees, 20 degrees, 30 degrees): TE, 0.88 milliseconds; TR, 2.65 milliseconds; slice thickness, 8.8 mm for 4 slices; acquisition matrix, 128 x 128; and acquisitions, 4. These data served to calculate 3D intrinsic longitudinal relaxation rate maps (R10) and magnetization (M0). Immediately after these images, the dynamic 3D T1-weighted gradient echo images were acquired with the same parameters and a constant alpha = 30 degrees, half Fourier, 1 acquisition, 64 frames, a time interval of 1.65 seconds between each frame, and a total duration of 105.6. Three milliliters of an albumin-binding blood pool contrast agent (0.25 mmol/mL gadofosveset trisodium, Vasovist, Bayer Schering Pharma AG, Berlin, Germany) was injected at a rate of 3 mL/s. Perfusion was calculated using the arterial input function from the aorta, which was extracted from the dynamic relaxation rate change maps and perfusion images were calculated on a voxel-by-voxel basis using a singular value decomposition. RESULTS: In 11 pigs, 4 different perfusion states were investigated sequentially. The reduced kidney perfusion measured by ultrasound highly correlated with total renal blood flow determined by DCE-MRI, P < 0.001. The correlation coefficient between both measurements was 0.843. Regional cortical and medullary renal flow was also highly correlated (r = 0.77/0.78, P < 0.001) with the degree of flow reduction. Perfusion values smaller than 50 mL/min/100 cm were overestimated by MRI, high perfusion values slightly underestimated. CONCLUSION: DCE-MRI using a blood pool contrast agent allows absolute quantification of total kidney perfusion as well as separate determination of cortical and medullary flow. The results show that our technique has sufficient accuracy and reproducibility to be transferred to the clinical setting.