Dataset Information

Impact of Pre- and Post-Processing Steps for Supervised Classification of Colorectal Cancer in Hyperspectral Images.

ABSTRACT:

Background

Recent studies have shown that hyperspectral imaging (HSI) combined with neural networks can detect colorectal cancer. Usually, different pre-processing techniques (e.g., wavelength selection and scaling, smoothing, denoising) are analyzed in detail to achieve a well-trained network. The impact of post-processing was studied less.

Methods

We tested the following methods: (1) Two pre-processing techniques (Standardization and Normalization), with (2) Two 3D-CNN models: Inception-based and RemoteSensing (RS)-based, with (3) Two post-processing algorithms based on median filter: one applies a median filter to a raw predictions map, the other applies the filter to the predictions map after adopting a discrimination threshold. These approaches were evaluated on a dataset that contains ex vivo hyperspectral (HS) colorectal cancer records of 56 patients.

Results

(1) Inception-based models perform better than RS-based, with the best results being 92% sensitivity and 94% specificity; (2) Inception-based models perform better with Normalization, RS-based with Standardization; (3) Our outcomes show that the post-processing step improves sensitivity and specificity by 6.6% in total. It was also found that both post-processing algorithms have the same effect, and this behavior was explained.

Conclusion

HSI combined with tissue classification algorithms is a promising diagnostic approach whose performance can be additionally improved by the application of the right combination of pre- and post-processing.

SUBMITTER: Tkachenko M

PROVIDER: S-EPMC10093280 | biostudies-literature | 2023 Apr

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Impact of Pre- and Post-Processing Steps for Supervised Classification of Colorectal Cancer in Hyperspectral Images.

Tkachenko Mariia M Chalopin Claire C Jansen-Winkeln Boris B Neumuth Thomas T Gockel Ines I Maktabi Marianne M

Cancers 20230405 7

<h4>Background</h4>Recent studies have shown that hyperspectral imaging (HSI) combined with neural networks can detect colorectal cancer. Usually, different pre-processing techniques (e.g., wavelength selection and scaling, smoothing, denoising) are analyzed in detail to achieve a well-trained network. The impact of post-processing was studied less.<h4>Methods</h4>We tested the following methods: (1) Two pre-processing techniques (Standardization and Normalization), with (2) Two 3D-CNN models: I ...[more]

PMID: 37046818

Similar Datasets

Project description:With recent technological advances in microscopy and image acquisition of tissue sections, further developments of tools are required for viewing, transforming, and analyzing the ever-increasing amounts of high-resolution data produced. In the field of neuroscience, histological images of whole rodent brain sections are commonly used for investigating brain connections as well as cellular and molecular organization in the normal and diseased brain, but present a problem for the typical neuroscientist with no or limited programming experience in terms of the pre- and post-processing steps needed for analysis. To meet this need we have designed Nutil, an open access and stand-alone executable software that enables automated transformations, post-processing, and analyses of 2D section images using multi-core processing (OpenMP). The software is written in C++ for efficiency, and provides the user with a clean and easy graphical user interface for specifying the input and output parameters. Nutil currently contains four separate tools: (1) A transformation toolchain named "Transform" that allows for rotation, mirroring and scaling, resizing, and renaming of very large tiled tiff images. (2) "TiffCreator" enables the generation of tiled TIFF images from other image formats such as PNG and JPEG. (3) A "Resize" tool completes the preprocessing toolset and allows downscaling of PNG and JPEG images with output in PNG format. (4) The fourth tool is a post-processing method called "Quantifier" that enables the quantification of segmented objects in the context of regions defined by brain atlas maps generated with the QuickNII software based on a 3D reference atlas (mouse or rat). The output consists of a set of report files, point cloud coordinate files for visualization in reference atlas space, and reference atlas images superimposed with color-coded objects. The Nutil software is made available by the Human Brain Project (https://www.humanbrainproject.eu) at https://www.nitrc.org/projects/nutil/.

Project description:Along with the global climate change, there is an increasing interest for its effect on phenological patterns such as start and end of the growing season. Scientific digital webcams are used for this purpose taking every day one or more images from the same natural motive showing for example trees or grassland sites. To derive phenological patterns from the webcam images, regions of interest are manually defined on these images by an expert and subsequently a time series of percentage greenness is derived and analyzed with respect to structural changes. While this standard approach leads to satisfying results and allows to determine dates of phenological change points, it is associated with a considerable amount of manual work and is therefore constrained to a limited number of webcams only. In particular, this forbids to apply the phenological analysis to a large network of publicly accessible webcams in order to capture spatial phenological variation. In order to be able to scale up the analysis to several hundreds or thousands of webcams, we propose and evaluate two automated alternatives for the definition of regions of interest, allowing for efficient analyses of webcam images. A semi-supervised approach selects pixels based on the correlation of the pixels' time series of percentage greenness with a few prototype pixels. An unsupervised approach clusters pixels based on scores of a singular value decomposition. We show for a scientific webcam that the resulting regions of interest are at least as informative as those chosen by an expert with the advantage that no manual action is required. Additionally, we show that the methods can even be applied to publicly available webcams accessed via the internet yielding interesting partitions of the analyzed images. Finally, we show that the methods are suitable for the intended big data applications by analyzing 13988 webcams from the AMOS database. All developed methods are implemented in the statistical software package R and publicly available in the R package phenofun. Executable example code is provided as supplementary material.

Project description:Surgery for brain cancer is a major problem in neurosurgery. The diffuse infiltration into the surrounding normal brain by these tumors makes their accurate identification by the naked eye difficult. Since surgery is the common treatment for brain cancer, an accurate radical resection of the tumor leads to improved survival rates for patients. However, the identification of the tumor boundaries during surgery is challenging. Hyperspectral imaging is a non-contact, non-ionizing and non-invasive technique suitable for medical diagnosis. This study presents the development of a novel classification method taking into account the spatial and spectral characteristics of the hyperspectral images to help neurosurgeons to accurately determine the tumor boundaries in surgical-time during the resection, avoiding excessive excision of normal tissue or unintentionally leaving residual tumor. The algorithm proposed in this study to approach an efficient solution consists of a hybrid framework that combines both supervised and unsupervised machine learning methods. Firstly, a supervised pixel-wise classification using a Support Vector Machine classifier is performed. The generated classification map is spatially homogenized using a one-band representation of the HS cube, employing the Fixed Reference t-Stochastic Neighbors Embedding dimensional reduction algorithm, and performing a K-Nearest Neighbors filtering. The information generated by the supervised stage is combined with a segmentation map obtained via unsupervised clustering employing a Hierarchical K-Means algorithm. The fusion is performed using a majority voting approach that associates each cluster with a certain class. To evaluate the proposed approach, five hyperspectral images of surface of the brain affected by glioblastoma tumor in vivo from five different patients have been used. The final classification maps obtained have been analyzed and validated by specialists. These preliminary results are promising, obtaining an accurate delineation of the tumor area.

Dataset Information

Impact of Pre- and Post-Processing Steps for Supervised Classification of Colorectal Cancer in Hyperspectral Images.

Background

Methods

Results

Conclusion

Publications

Impact of Pre- and Post-Processing Steps for Supervised Classification of Colorectal Cancer in Hyperspectral Images.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets