On the different aspects of deep learning in bioimaging based diagnostics

In fifteen words or less, explain the significance of this contribution (Novel Aspect).: Research on deep learning to leverage bioimaging for diagnostics

Abstract Text:

Releasing the burden and improving the quality of healthcare has become an urgent need of society. Bioimaging techniques are shown promise in this respect. Therein optical signals are acquired from biological samples (e.g., cells, biofluid, tissues) and translated into biological/clinical knowledge that are relevant to diagnostics. This has been largely supported by deep learning nowadays. Herein, we present a 3D neural network for Raman-imaging-based colorectal cancer diagnosis. By simultaneously employing spatial and spectral information, the 3D network led to more stable prediction and reduced the computation burden compared to 1D and 2D convolutional neural networks [1]. To better deal with the scarcity of training samples in diagnostic tasks, we systematically investigated the different transfer learning strategies with colorectal cancer diagnosis as example data [2]. The sample size required for a satisfying diagnosis was estimated with our established approach [3]. Moreover, we employed different data augmentation approaches to improve the performance of prediction with lower demand for training samples. With all studies presented, we hope to benefit both methodology and knowledge on deep learning and further to leverage the potential of bioimaging in diagnostics.

We highly acknowledge the project TELEGRAFT (101057673) with the funding from the European Union’s 9th Framework Program Horizon Europe. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the Health and Digital Executive Agency. Neither the European Union nor the granting authority can be held responsible for them.

[1] R. Luo, S. Guo, J. Hniopek, T. Bocklitz, Analytical Chemistry, 97 (14), 7729–7737.
[2] R. Luo, T Bocklitz, Informatics in Medicine Unlocked, 2023, 40, 101292
[3] R. Luo, S. Guo, T Bocklitz, ICPRAM, 841-851.