DiviK: divisive intelligent K-means for hands-free unsupervised clustering in big biological data.

BACKGROUND: Investigating molecular heterogeneity provides insights into tumour origin and metabolomics. The increasing amount of data gathered makes manual analyses infeasible-therefore, automated unsupervised learning approaches are utilised for discovering tissue heterogeneity. However, automated analyses require experience setting the algorithms' hyperparameters and expert knowledge about the analysed biological processes. Moreover, feature engineering is needed to obtain valuable results because of the numerous features measured. RESULTS: We propose DiviK: a scalable stepwise algorithm with local data-driven feature space adaptation for segmenting high-dimensional datasets. The algorithm is compared to the optional solutions (regular k-means, spatial and spectral approaches) combined with different feature engineering techniques (None, PCA, EXIMS, UMAP, Neural Ions). Three quality indices: Dice Index, Rand Index and EXIMS score, focusing on the overall composition of the clustering, coverage of the tumour region and spatial cluster consistency, are used to assess the quality of unsupervised analyses. Algorithms were validated on mass spectrometry imaging (MSI) datasets-2D human cancer tissue samples and 3D mouse kidney images. DiviK algorithm performed the best among the four clustering algorithms compared (overall quality score 1.24, 0.58 and 162 for d(0, 0, 0), d(1, 1, 1) and the sum of ranks, respectively), with spectral clustering being mostly second. Feature engineering techniques impact the overall clustering results less than the algorithms themselves (partial [Formula: see text] effect size: 0.141 versus 0.345, Kendall's concordance index: 0.424 versus 0.138 for d(0, 0, 0)). CONCLUSIONS: DiviK could be the default choice in the exploration of MSI data. Thanks to its unique, GMM-based local optimisation of the feature space and deglomerative schema, DiviK results do not strongly depend on the feature engineering technique applied and can reveal the hidden structure in a tissue sample. Additionally, DiviK shows high scalability, and it can process at once the big omics data with more than 1.5 mln instances and a few thousand features. Finally, due to its simplicity, DiviK is easily generalisable to an even more flexible framework. Therefore, it is helpful for other -omics data (as single cell spatial transcriptomic) or tabular data in general (including medical images after appropriate embedding). A generic implementation is freely available under Apache 2.0 license at https://github.com/gmrukwa/divik .

Classification of Thyroid Tumors Based on Mass Spectrometry Imaging of Tissue Microarrays; a Single-Pixel Approach.

The primary diagnosis of thyroid tumors based on histopathological patterns can be ambiguous in some cases, so proper classification of thyroid diseases might be improved if molecular biomarkers support cytological and histological assessment. In this work, tissue microarrays representative for major types of thyroid malignancies-papillary thyroid cancer (classical and follicular variant), follicular thyroid cancer, anaplastic thyroid cancer, and medullary thyroid cancer-and benign thyroid follicular adenoma and normal thyroid were analyzed by mass spectrometry imaging (MSI), and then different computation approaches were implemented to test the suitability of the registered profiles of tryptic peptides for tumor classification. Molecular similarity among all seven types of thyroid specimens was estimated, and multicomponent classifiers were built for sample classification using individual MSI spectra that corresponded to small clusters of cells. Moreover, MSI components showing the most significant differences in abundance between the compared types of tissues detected and their putative identity were established by annotation with fragments of proteins identified by liquid chromatography-tandem mass spectrometry in corresponding tissue lysates. In general, high accuracy of sample classification was associated with low inter-tissue similarity index and a high number of components with significant differences in abundance between the tissues. Particularly, high molecular similarity was noted between three types of tumors with follicular morphology (adenoma, follicular cancer, and follicular variant of papillary cancer), whose differentiation represented the major classification problem in our dataset. However, low level of the intra-tissue heterogeneity increased the accuracy of classification despite high inter-tissue similarity (which was exemplified by normal thyroid and benign adenoma). We compared classifiers based on all detected MSI components (n = 1536) and the subset of the most abundant components (n = 147). Despite relatively higher contribution of components with significantly different abundance and lower overall inter-tissue similarity in the latter case, the precision of classification was generally higher using all MSI components. Moreover, the classification model based on individual spectra (a single-pixel approach) outperformed the model based on mean spectra of tissue cores. Our result confirmed the high feasibility of MSI-based approaches to multi-class detection of cancer types and proved the good performance of sample classification based on individual spectra (molecular image pixels) that overcame problems related to small amounts of heterogeneous material, which limit the applicability of classical proteomics.

Fully-automated deep learning-powered system for DCE-MRI analysis of brain tumors.

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) plays an important role in diagnosis and grading of brain tumors. Although manual DCE biomarker extraction algorithms boost the diagnostic yield of DCE-MRI by providing quantitative information on tumor prognosis and prediction, they are time-consuming and prone to human errors. In this paper, we propose a fully-automated, end-to-end system for DCE-MRI analysis of brain tumors. Our deep learning-powered technique does not require any user interaction, it yields reproducible results, and it is rigorously validated against benchmark and clinical data. Also, we introduce a cubic model of the vascular input function used for pharmacokinetic modeling which significantly decreases the fitting error when compared with the state of the art, alongside a real-time algorithm for determination of the vascular input region. An extensive experimental study, backed up with statistical tests, showed that our system delivers state-of-the-art results while requiring less than 3 min to process an entire input DCE-MRI study using a single GPU.

Segmenting brain tumors from FLAIR MRI using fully convolutional neural networks.

BACKGROUND AND OBJECTIVE: Magnetic resonance imaging (MRI) is an indispensable tool in diagnosing brain-tumor patients. Automated tumor segmentation is being widely researched to accelerate the MRI analysis and allow clinicians to precisely plan treatment-accurate delineation of brain tumors is a critical step in assessing their volume, shape, boundaries, and other characteristics. However, it is still a very challenging task due to inherent MR data characteristics and high variability, e.g., in tumor sizes or shapes. We present a new deep learning approach for accurate brain tumor segmentation which can be trained from small and heterogeneous datasets annotated by a human reader (providing high-quality ground-truth segmentation is very costly in practice). METHODS: In this paper, we present a new deep learning technique for segmenting brain tumors from fluid attenuation inversion recovery MRI. Our technique exploits fully convolutional neural networks, and it is equipped with a battery of augmentation techniques that make the algorithm robust against low data quality, and heterogeneity of small training sets. We train our models using only positive (tumorous) examples, due to the limited amount of available data. RESULTS: Our algorithm was tested on a set of stage II-IV brain-tumor patients (image data collected using MAGNETOM Prisma 3T, Siemens). Rigorous experiments, backed up with statistical tests, revealed that our approach outperforms the state-of-the-art approach (utilizing hand-crafted features) in terms of segmentation accuracy, offers very fast training and instant segmentation (analysis of an image takes less than a second). Building our deep model is 1.3 times faster compared with extracting features for extremely randomized trees, and this training time can be controlled. Finally, we showed that too aggressive data augmentation may lead to deteriorated performance of the model, especially in the fixed-budget training (with maximum numbers of training epochs). CONCLUSIONS: Our method yields the better performance when compared with the state of the art method which utilizes hand-crafted features. In addition, our deep network can be effectively applied to difficult (small, imbalanced, and heterogeneous) datasets, offers controllable training time, and infers in real-time.

Discrimination of normal oral mucosa from oral cancer by mass spectrometry imaging of proteins and lipids.

Identification of biomarkers for molecular classification of cancer and for differentiation between cancerous and normal epithelium remains a vital issue in the field of head and neck cancer. Here we aimed to compare the ability of proteome and lipidome components to discriminate oral cancer from normal mucosa. Tissue specimens including squamous cell cancer and normal epithelium were analyzed by MALDI mass spectrometry imaging. Two molecular domains of tissue components were imaged in serial sections-peptides (resulting from trypsin-processed proteins) and lipids (primarily zwitterionic phospholipids), then regions of interest corresponding to cancer and normal epithelium were compared. Heterogeneity of cancer regions was higher than the heterogeneity of normal epithelium, and the distribution of peptide components was more heterogeneous than the distribution of lipid components. Moreover, there were more peptide components than lipid components that showed significantly different abundance between cancer and normal epithelium (median of the Cohen's effect was 0.49 and 0.31 in case of peptide and lipid components, respectively). Multicomponent cancer classifier was tested (vs. normal epithelium) using tissue specimens from three patients and then validated with a tissue specimen from the fourth patient. Peptide-based signature and lipid-based signature allowed cancer classification with a weighted accuracy of 0.85 and 0.69, respectively. Nevertheless, both classifiers had very high precision (0.98 and 0.94, respectively). We concluded that though molecular differences between cancerous and normal mucosa were higher in the proteome domain than in the analyzed lipidome subdomain, imaging of lipidome components also enabled discrimination of oral cancer and normal epithelium. Therefore, both cancer proteome and lipidome are promising sources of biomarkers of oral malignancies.

Molecular profiles of thyroid cancer subtypes: Classification based on features of tissue revealed by mass spectrometry imaging.

Determination of the specific type of thyroid cancer is crucial for the prognosis and selection of treatment of this malignancy. However, in some cases appropriate classification is not possible based on histopathological features only, and it might be supported by molecular biomarkers. Here we aimed to characterize molecular profiles of different thyroid malignancies using mass spectrometry imaging (MSI) which enables the direct annotation of molecular features with morphological pictures of an analyzed tissue. Fifteen formalin-fixed paraffin-embedded tissue specimens corresponding to five major types of thyroid cancer were analyzed by MALDI-MSI after in-situ trypsin digestion, and the possibility of classification based on the results of unsupervised segmentation of MALDI images was tested. Novel method of semi-supervised detection of the cancer region of interest (ROI) was implemented. We found strong separation of medullary cancer from malignancies derived from thyroid epithelium, and separation of anaplastic cancer from differentiated cancers. Reliable classification of medullary and anaplastic cancers using an approach based on automated detection of cancer ROI was validated with independent samples. Moreover, extraction of spectra from tumor areas allowed the detection of molecular components that differentiated follicular cancer and two variants of papillary cancer (classical and follicular). We concluded that MALDI-MSI approach is a promising strategy in the search for biomarkers supporting classification of thyroid malignant tumors. This article is part of a Special Issue entitled: MALDI Imaging, edited by Dr. Corinna Henkel and Prof. Peter Hoffmann.

Detection of molecular signatures of oral squamous cell carcinoma and normal epithelium - application of a novel methodology for unsupervised segmentation of imaging mass spectrometry data.

Intra-tumor heterogeneity is a vivid problem of molecular oncology that could be addressed by imaging mass spectrometry. Here we aimed to assess molecular heterogeneity of oral squamous cell carcinoma and to detect signatures discriminating normal and cancerous epithelium. Tryptic peptides were analyzed by MALDI-IMS in tissue specimens from five patients with oral cancer. Novel algorithm of IMS data analysis was developed and implemented, which included Gaussian mixture modeling for detection of spectral components and iterative k-means algorithm for unsupervised spectra clustering performed in domain reduced to a subset of the most dispersed components. About 4% of the detected peptides showed significantly different abundances between normal epithelium and tumor, and could be considered as a molecular signature of oral cancer. Moreover, unsupervised clustering revealed two major sub-regions within expert-defined tumor areas. One of them showed molecular similarity with histologically normal epithelium. The other one showed similarity with connective tissue, yet was markedly different from normal epithelium. Pathologist's re-inspection of tissue specimens confirmed distinct features in both tumor sub-regions: foci of actual cancer cells or cancer microenvironment-related cells prevailed in corresponding areas. Hence, molecular differences detected during automated segmentation of IMS data had an apparent reflection in real structures present in tumor.