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.

Frequency-invariant representation of interaural time differences in mammals.

Interaural time differences (ITDs) are the major cue for localizing low-frequency sounds. The activity of neuronal populations in the brainstem encodes ITDs with an exquisite temporal acuity of about 10 µs. The response of single neurons, however, also changes with other stimulus properties like the spectral composition of sound. The influence of stimulus frequency is very different across neurons and thus it is unclear how ITDs are encoded independently of stimulus frequency by populations of neurons. Here we fitted a statistical model to single-cell rate responses of the dorsal nucleus of the lateral lemniscus. The model was used to evaluate the impact of single-cell response characteristics on the frequency-invariant mutual information between rate response and ITD. We found a rough correspondence between the measured cell characteristics and those predicted by computing mutual information. Furthermore, we studied two readout mechanisms, a linear classifier and a two-channel rate difference decoder. The latter turned out to be better suited to decode the population patterns obtained from the fitted model.

A two-dimensional cochlear fluid model based on conformal mapping.

Using conformal mapping, fluid motion inside the cochlear duct is derived from fluid motion in an infinite half plane. The cochlear duct is represented by a two-dimensional half-open box. Motion of the cochlear fluid creates a force acting on the cochlear partition, modeled by damped oscillators. The resulting equation is one-dimensional, more realistic, and can be handled more easily than existing ones derived by the method of images, making it useful for fast computations of physically plausible cochlear responses. Solving the equation of motion numerically, its ability to reproduce the essential features of cochlear partition motion is demonstrated. Because fluid coupling can be changed independently of any other physical parameter in this model, it allows the significance of hydrodynamic coupling of the cochlear partition to itself to be quantitatively studied. For the model parameters chosen, as hydrodynamic coupling is increased, the simple resonant frequency response becomes increasingly asymmetric. The stronger the hydrodynamic coupling is, the slower the velocity of the resulting traveling wave at the low frequency side is. The model's simplicity and straightforward mathematics make it useful for evaluating more complicated models and for education in hydrodynamics and biophysics of hearing.