For more than a decade, machine learning (ML) and deep learning (DL) techniques have been a mainstay in the toolset for the analysis of large amounts of weakly correlated or high-dimensional data. As new technologies for detecting and measuring biochemical markers from bodily fluid samples (e.g., microfluidics and labs-on-a-chip) revolutionise the industry of diagnostics and precision medicine, the heterogeneity and complexity of the acquired data present a growing challenge to their interpretation and usage. In this chapter, we attempt to review the state of ML and DL fields as applied to the analysis of liquid biopsy data and summarise the available corpus of techniques and methodologies.

In the global cancer statistics, hepatocellular carcinoma (HCC) ranges sixth by incidence and second by oncological mortality. The risk factors comprise hepatitis B and C virus infection, non-alcoholic steatohepatitis, as well as long-lasting peroral exposure to alcohol or aflatoxins. Liver cirrhosis is the most important single predisposing factor. Ultrasonography once per 6 months is recommended for surveillance in cirrhotic patients. Computed tomography (CT) and magnetic resonance imaging (MRI) represent the gold standard of noninvasive diagnostics while core biopsy and/or immunohistochemistry (IHC) are indicated for controversial and non-cirrhotic HCC cases. Molecular classification is under development. At present, classics of HCC diagnostics is based on evaluation of risk factors, surveillance in cirrhotic patients, preference for CT or MRI-confirmed non-invasive diagnosis and biopsy proof in equivocal cases. Diffusion-weighted imaging and hepatobiliary phase contrasting represent significant recent developments in MRI. Contrast-enhanced ultrasonography is recommended by some but not all guidelines. Positron emission tomography is advocated before liver transplantation to detect extrahepatic metastases but has limited role in the initial diagnostic evaluation of liver nodule. Innovations are expected in the field of molecular diagnostics, including IHC panels and novel antigens, e.g. clathrin and bile salt export pump protein, and development of molecular classification.

Hepatocellular carcinoma (HCC) is an aggressive tumour associated with dismal prognosis. To improve the outcome, early diagnostics is important. At present, classical HCC diagnostics is based on evaluation of risk factors, surveillance in cirrhotic patients, preference for non-invasive diagnosis by computed tomography or magnetic resonance imaging and biopsy confirmation in controversial cases. However, ambiguous radiological presentation, biopsy-related complications or insufficient representation of the pathology in the tissue core are well-known problems. Panel assessment of microRNAs has diagnostic and prognostic value; thus, in future, microRNA-based liquid biopsy could partially reduce the need for core biopsies. Systemic inflammatory reaction (SIR), characterised mainly by neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio and Glasgow prognostic score, may have prognostic value and can be incorporated in criteria for certain treatment approaches, e.g., becoming an adjunct to Milan criteria. Thus, innovations in HCC diagnostics are expected in the field of miRNA-based liquid biopsy for diagnosis/prognosis and SIR for prognosis/selection of treatment.

Thyroid nodules are frequent in general population, found in 3.7-7% of people by palpation and 42-67% by ultrasonography (US). The differential diagnosis ranges from papillary (PC), follicular (FC) and medullary (MC) carcinomas to follicular adenoma (FA) and colloid goitre. Cancer risk in thyroid nodules varies: 5% in masses found by palpation, 1.6-15% by US, 3.9-11.3% by computed tomography (CT), 5-6% by magnetic resonance imaging (MRI) and 30-50% by positron emission tomography (PET). The final diagnosis depends on fine needle aspiration (FNA) findings and histopathology. The recent WHO classification (2017) is based on classic morphology, including assessment of invasion and nuclei. New entities are defined to designate tumours with doubtful invasion or controversial nuclear features. By immunohistochemistry, PC expresses HBME-1, TROP-2, CITED1 and CK19. Notably, PC can stain for CD20. MC is recognised by neuroendocrine differentiation. To distinguish FA vs. FC, evaluation of HBME-1, p27 and galectin has been suggested. Regarding miRNAs, miR-146b, miR-222, miR-221 and miR-181b are upregulated, while miR-145, miR-451, miR-613 and miR-137 are downregulated in PC. FC features downregulated miR-199a-5p and upregulated miR-197 and miR-346. In MC, miR-21 and miR-129-5p are downregulated. In addition, increased systemic inflammatory reaction can be poor prognostic factor in thyroid cancer. The aim of this chapter is to review classic and innovative histopathology of thyroid nodules for diagnostic pathology practice and research in multidisciplinary thyroid teams.

Thyroid cancer is a comparatively rare tumor, which affects 1-5% of women and approximately 2% of men, although it is the most common endocrine malignancy worldwide. Furthermore, the incidence of thyroid cancer has been increasing remarkably in the last decades. Currently, diagnosis of thyroid cancer mainly is based on cytological criteria. Although fine needle aspiration is a minimally invasive procedure, complications can occur. Correct diagnosis is mandatory to select patients for surgical intervention and to determine appropriate extent of operation. Overdiagnosis and the associated unnecessary surgery should be avoided as it might also lead to complications. Therefore it is important to practice noninvasive methods not only for early diagnosis of thyroid cancer but also for estimation of prognosis. Liquid biopsy is a promising, noninvasive method that can provide detection of circulating tumor cells (CTCs) as well as circulating nucleic acids such as DNA, mRNA, and microRNA in a blood sample. The aim of the chapter is to highlight the efficacy of liquid biopsy for diagnosis and prognosis of thyroid cancer. The chapter will represent a comprehensive literature review based on recent PubMed publications (mainly 2012-2018).