Introduction
Hepatic neoplasms are categorically classified into primary tumours originating within the liver and secondary tumours arising from metastatic dissemination of malignancies from other primary sites. Hepatocellular carcinoma (HCC) is a leading global cancer (comprising 75–85% of cases) and the most common primary liver malignancy. It is ranked second in cancer-related deaths and is primarily linked to chronic liver diseases caused by hepatitis B or C, consumption of aflatoxin-contaminated foods, excessive alcohol consumption, obesity, type 2 diabetes, tobacco smoking, and metabolic-associated fatty liver disease (MAFLD) [1–5]. HCC incidence varies geographically, with higher rates in HBV-endemic regions. In most cases, HCC develops in the context of cirrhosis. Cholangiocarcinoma (CCA) (10–15%) is the second most common primary liver cancer [2–7]. It arises from the epithelial cells of the intrahepatic and extrahepatic bile ducts and is classified based on its anatomical location in the biliary tree [7–9]. Combined hepatocellular-cholangiocarcinoma (cHCC-CCA) is a primary malignancy of the liver displaying characteristics of both hepatic and biliary differentiation. This tumour, classified as a “biphenotypic” neoplasm, is a seldom observed phenomenon, accounting for fewer than 5% of all primary liver cancers [7]. Epithelioid haemangioendothelioma (EH) represents a rare low-grade malignant liver tumour, typically asymptomatic at detection [10, 11]. Liver angiosarcoma, originating from endothelial cells, is exceptionally rare and highly aggressive [12–14]. Malignant fibrous histiocytoma (MFH) is an uncommon liver tumour, with extremities being the most common site [12, 13]. Undifferentiated embryonal sarcoma (UES) is a highly malignant liver neoplasm typically found in children aged 6 to 10 years [12, 13, 15]. Nested stromal epithelial tumour (NSET), a recently identified primary neoplasm, is extremely rare in the liver [16]. Primary hepatic neuroendocrine tumours (PHNT) constitute a mere 0.3% of all neuroendocrine tumours and are exceptionally rare [17]. The liver serves as a frequent site for metastasis, primarily attributed to its distinctive cellular and architectural composition that provides a conducive environment for tumour cells [18, 19]. Various malignancies originating from the gastrointestinal tract, breast, prostate, neuroendocrine system, and sarcomas exhibit a propensity to metastasise to the liver. Notably, in the United States, liver metastases surpass primary liver tumours in prevalence [18, 20, 21]. Metastatic diseases, accounting for over 90% of solid tumour-related mortalities, exert a significant impact on the healthcare system, leading to escalated costs, heightened spending, and increased demand for healthcare resources [22, 23]. The existing body of literature predominantly focuses on specific cancer subtypes, particularly colorectal cancer, while overlooking other cancers with liver metastatic potential, such as pancreatic, breast, and prostate cancers [24–27]. Historical data suggest that individuals with liver metastases have a poorer prognosis compared to those without this condition. However, there is a lack of direct quantification of this disparity, and few contemporary studies that provide supporting evidence for these observations [27, 28]. This study aimed to comprehensively explore the profound impact of artificial intelligence (AI) and image processing in the context of liver cancer, with a specific focus on their pivotal roles in the domains of diagnosis, treatment guidance, and prognosis. Through an in-depth narrative review, we aim to elucidate the various applications of AI and image processing in enhancing the understanding and management of liver cancer, highlighting their potential to revolutionise current practices in oncology. Two independent reviewers screened the studies, resolving any discrepancies through discussion.
AI and medical applications
AI involves the creation of computer algorithms that perform tasks typically associated with human intelligence. AI encompasses a wide spectrum of technological advancements facilitating the simulation of human cognitive abilities by robots and computers [29, 30]. This term is widely used to encompass various forms of learning, including machine learning, representation learning, deep learning, natural language processing, and large language models. There exists a wide range of potential applications for large language model (LLM) technology in the field of medicine, benefiting both patients and healthcare providers [31]. The overarching goal of AI in the field of medicine is to utilise these technologies to extract pertinent information from well-organised databases and aid in affecting clinical decision-making [32]. AI technologies have multiple functions, such as assisting in diagnoses and selecting therapies, predicting risks, stratifying diseases, minimising medical errors, and enhancing productivity [33, 34].
Radiomics in liver lesion diagnosis
Radiological diagnosis of focal liver lesions relies on the subjective visual assessment of ultrasound (US), computed tomography (CT), or magnetic resonance imaging (MRI). However, the heterogeneity of these lesions poses challenges in accurate classification. Therefore, radiomics on hepatic lesions has gained interest. Numerous studies have demonstrated the feasibility of liver tumour classification using radiomics across various imaging modalities [35–42].
Zhao et al. reported a computer-aided approach to distinguish 6 types of hepatic lesions based on radiomics features in unenhanced CT scans, achieving an overall accuracy of 88% and 76% in training and testing sets, respectively [35]. Oyama et al. utilised texture analysis and topological data analysis on T1-weighted MR images to differentiate between HCC, metastatic tumours, and HHs, reaching an accuracy of 92% [36]. In differentiating between benign and malignant hypervascular hepatocellular lesions, radiomics approaches have shown promising results [37]. Nomograms combining radiomics scores and clinical factors exhibited good differentiation capability in preoperative assessments. Radiomics analyses with clinical factors were also effective in discriminating between focal nodular hyperplasia (FNH) and HCC, as well as FNH and hepatocellular adenoma (HCA) [37, 38]. Recent advancements in radiomics including a study by Wang et al. with the use of PyRadiomics to extract 1316 features from MR images for distinguishing between combined hepatocellular cholangiocarcinoma, HCC, and cholangiocarcinoma. Their model, incorporating high-order features, demonstrated a 10% improvement in classification compared to low-order image features. These findings underscore the potential of radiomics in enhancing the precision of liver lesion diagnosis and classification [41]. Chu et al. analysed 203 intrahepatic cholangiocarcinomas (ICCs) and developed a predictive model for avoidable resections. The radiomic model outperformed the clinical model, achieving a sensitivity of 0.846
and a specificity of 0.771 in the validation cohort [43]. In a study by Quin et al., a multilevel model incorporating clinicopathology, molecular pathology, and radiology predicted early recurrence after curative intent surgery [44]. Analysing 18,120 radiomic features from CT studies and 48 clinical features, the radiomics-based model demonstrated superior performance to conventional staging systems, offering prognostic value for postoperative management. Hao et al. devised a non-invasive CT-based radiomics analysis model for predicting early recurrence in 177 ICC patients. Employing Max-Relevance Min-Redundancy (MRMR) and Gradient Boosting Machine (GBM) on stable radiomic models, the model achieved high AUCs of 0.802 and 0.781 in the training and testing sets, respectively [45]. In total, radiomics, corresponding to the extraction and analysis of quantitative features from medical images, has emerged as a promising field in radiology [46]. It can be coupled with AI, enhancing diagnostic accuracy and providing more valuable insights for personalised medicine [47]. Indeed, the last decade has seen increased development of computerised tools that convert images into quantitative mineable data (radiomics) and their subsequent analyses with AI [48]. Radiomics, which utilises computers to extract a large amount of information from different types of images, and forms various quantifiable features aided by AI algorithms that select relevant features to build models for predicting the outcomes of clinical problems. However, while initial studies looking at radiomics have been very promising, there has been poor standardisation and generalisation of radiomic results, limiting the translation of this approach into clinical practice. This is particularly evident in issues related to data quality control, repeatability, reproducibility, generalisability of results, and concerns about model overfitting. To address these limits, Hu et al. propose that future radiomic research should be assessed via the radiomics quality score established by Lambin et al. [49, 50]. Awareness of their potential is the prerequisite for improving research standards and encouraging their adoption by the scientific community [51].
The role of image processing and machine learning in liver cancer
Machine learning (ML), a closely linked discipline within the field of AI, refers to the overarching concept denoting a machine’s capability to learn and subsequently enhance its patterns and models for analytical purposes [52, 53]. In the field of image processing, an essential step is referred to as “image smoothing”, which aims to enhance the quality of the image by reducing noise and distortions. This stage plays a crucial role in facilitating the extraction and classification of image features. Particularly in the analysis of biological images, it becomes imperative to utilise an appropriate filtering methodology. To achieve exceptional outcomes, it becomes necessary to carefully select a liver tumour image denoising algorithm that is highly effective. Additionally, the employment of segmentation algorithms holds promise in effectively removing undesirable elements from liver images. Further investigation and advancement in research and development are needed in the areas of denoising, segmentation, feature selection, and prediction algorithm selection as they pertain to liver tumour imagery [54]. Chlebus et al. performed feature extraction on CT imaging data of liver cancer and utilised these features as training parameters to construct an artificial neural network (ANN)-assisted diagnosis model. This ANN was simulated and employed for liver cancer diagnosis in CT images. The results demonstrated notable enhancements in the sensitivity and specificity of liver cancer diagnosis [55].
AI and histopathological applications in liver specimens
In recent years, advancements in AI have significantly contributed to its widespread application within the medical domain. Specifically, deep learning algorithms have emerged as potent tools in machine learning, leveraging deep neural network architectures [56, 57]. These networks have found utilisation in diverse tasks such as computer vision and natural language processing [58]. Of particular significance is their remarkable application in medical image analysis and timely detection of various diseases [59–62]. Yu-shiang et al. employed a binary classifier based on GoogLeNet (Inception-V1) to classify histopathology images of hepatocellular carcinoma (HCC). The classifier demonstrated an accuracy of 91.37% (±2.49), a sensitivity of 92.16% (±4.93), and a specificity of 90.57% (±2.54) in the classification of HCC [62]. Lu et al. utilised pre-trained CNN models, namely VGG 16, Inception V3, and ResNet 50, to effectively differentiate between normal and cancer samples based on extracted HCC histopathological images. Nevertheless, one limitation of the study is the absence of multiple classification sample categories [63]. The integration of a multiphoton microscope with a deep learning algorithm has been empirically validated to effectively categorise hepatocellular carcinoma (HCC) differentiation, presenting a groundbreaking approach to computer-aided diagnosis [64]. This study aimed to assess the performance of 5 specific deep learning models – VGG16, ResNet50, ResNet_CBAM, SENet, and SKNet – in the context of classification experiments [65, 66]. The VGG16 network is characterised by its simplicity and a relatively small number of hyperparameters. It is primarily focused on constructing a convolutional layer using a 3 × 3 filter with a stride of 1, wherein the padding parameter is considered. This architectural design of the VGG network simplifies the neural network structure. However, one of the drawbacks of the VGG network is its large number of parameters, which leads to significant resource consumption during training processes [67, 68]. CBAM is a straightforward and efficient module used in convolutional neural networks. It enhanced the representation ability of intermediate feature maps within the network. Additionally, CBAM is designed as a versatile and integrated module that can be seamlessly incorporated into any end-to-end neural network architecture [69]. The concept underlying the SENet is characterised by its simplicity, enabling straightforward integration with pre-existing network structures. The SENet model introduces a novel block structure known as the SE block, wherein each feature layer is employed within the squeeze compression model architecture, employing excitation to effectively capture feature channel dependencies [70, 71]. SKNet and SENet are modules that can be seamlessly incorporated into a neural network architecture. These modules employ the attention mechanism to merge feature maps obtained from convolutional kernels of varying sizes. The size of the attention is determined by the deterministic information extracted by these diverse convolutional kernels [72]. The SKNet model incorporates a soft attention mechanism into the network architecture, enabling the network to gather information from various receptive fields. This enhances the generalisation ability of the network, resulting in a network structure that performs better on different tasks and datasets [73].
AI in the treatment of liver cancer
To overcome limitations commonly associated with performing liver resections, various definitions and risk scores exist for evaluating the complexity of laparoscopic liver resections, with the IWATE criteria being the prevailing choice in current practice [74]. Operating on tumours greater than 10 cm poses significant challenges, often pushing the boundaries of what is technically resectable. Trocar placement, liver lobe mobilisation, and tumour perforation due to shear forces are among the intraoperative difficulties that may arise [75]. Filmann et al. reported that the mortality rate following liver resection exhibits an incremental rise commensurate with the extent and necessity of biliary reconstruction, potentially reaching as high as 25% [76]. Augmented reality (AR) was used as an alternative tool, which requires the implementation of robotic hepatic surgery technologies for guidance in challenging anatomical locations. Despite the enhanced depth perception provided by the 3D visualisation and the improved video resolution and magnification, the robotic system still presents the disadvantages inherent to a minimally invasive approach in comparison to an open hepatectomy [77, 78]. From an oncological perspective, augmented reality (AR) technology in combination with AI-assisted programs plays a significant role in tumour localisation and guiding parenchymal transection during surgical procedures. It proved effective in preoperative planning by providing a 3D rendering of the tumour and its surrounding structures. Intraoperatively, AI in combination with AR allows precise targeting of the lesion and resection margins, enhancing the accuracy of surgical resection. AR offers advanced visualisation tools and assists in optimising surgical strategies for improved outcomes in cancer treatment [77–82]. Tactile feedback provided by the sense of touch assists the operator in accurate positioning about various intrahepatic landmarks. Notably, in the first order pedicles, arteries, veins, and biliary structures exhibit a thickened fibrotic sheath, and the lack of sensation in robotic surgery may lead to surgeon disorientation during dissection, potentially resulting in vascular injuries. Moreover, the presence of lesions in vital regions such as the hepatic confluence may render robotic dissection more intricate [81]. Heat-based (microwave, radiofrequency) and other types of transhepatic ablative procedures are considered alternative treatments for small hepatic lesions (< 3 cm ideally), offering similar oncological outcomes to liver resections, with fewer postoperative complications [83, 84]. Intraoperative ultrasound (US) guidance is typically utilised to localise the lesion. However, the US lacks a three-dimensional perception, and the ablation process itself can produce artifacts that obscure the target, thereby affecting the efficacy of the treatment. To address these limitations, some studies have proposed the use of an AR interface in preclinical studies, which would allow for better planning of the procedure and guide needle placement through a cooperative system involving a surgeon robot [85, 86]. Over the last 2 decades, there has been a notable and continuous escalation in the quantity and intricacy of robotic-guided interventions [87, 88]. Originally trailed in the late 1980s within the realm of neurosurgery, these systems subsequently underwent development for application in interventional radiology [89, 90]. Gradually, they have become a useful tool for performing biopsies and tumour ablations, proving particularly invaluable in instances that would otherwise necessitate a high level of expertise [91, 92]. In a 2023 systematic review of 429 percutaneous liver thermal ablations and 57 liver biopsies, all robot-guided, Bodard et al. demonstrated that the mean deviation of the probes was reduced by 30% and that robot-assisted interventions required 40% fewer readjustments [93]. Additionally, robotic systems reduced operating time, ranging from 15% to 25%. Finally, the average radiation dose delivered to the patient and operator was reduced by 50% compared to manual procedures. As of today, a few robotic devices have received FDA and/or CE approval, including the MAXIO, EPIONE, ROBIO-EX, AcuBot, and ACE robotic systems [93]. These procedures have garnered significant attention in the context of liver interventions, attributed to the elevated prevalence of liver diseases, the anatomical challenges of the liver, and the escalating demand for minimally invasive methodologies [94, 95]. In the domain of liver procedures, ensuring a secure trajectory to reach the target without causing harm to critical structures assumes paramount importance. This is especially critical for tumour ablation procedures such as microwave ablation (MWA), radiofrequency ablation (RFA), and irreversible electroporation (IRE), where a meticulous and efficient workflow is imperative for ensuring both the safety and efficacy of the treatment [96, 97]. The quantity of needle insertions, coupled with considerations such as depth, location, and tumour size, profoundly influences the safety of the procedure for the patient and subsequently impacts treatment outcomes [97, 98]. Advanced software improves post-operative ablation margin assessment, enhancing precision in procedures like microwave and radiofrequency ablation [98]. This leads to more effective treatment, reducing the risk of residual tumour cells and positively impacting long-term patient outcomes. Post-operative failure to meticulously account for these parameters may increase the risk of complications or tumour cell seeding, particularly in the context of challenging tumour locations such as the caudate lobe or the hepatic dome. Neglecting these parameters may elevate the risk of complications or the seeding of tumour cells [95–98]. In the context of a liver transplantation program, the primary constraint in the provision of medical care currently resides in the limited availability of organs. The United Network for Organ Sharing (UNOS) has conducted a survey that revealed a notable reduction of approximately 20% in the number of patients who remain eligible for liver transplantation [99]. The initial endeavour to guide the allocation of organs utilising donor information was carried out by Feng et al., who introduced the quantitative donor risk index [100]. This approach involved employing a Cox regression model to prognosticate graft failure solely based on donor characteristics [100]. Various strategies, including machine learning, are being explored to address the imbalance between the number of individuals in need of liver transplants and the limited availability of donated organs. Pérez-Ortiz et al. conducted a study employing ordinal regression and support vector machine techniques to develop a model that could be utilised alongside the Model for End-Stage Liver Disease (MELD) score. The aim was to allocate organs to those patients ranked highest on the waiting list according to their MELD scores, while also considering their likelihood of improved survival outcomes [101].
AI’s predictability of treatment response
Surgical resection is a curative treatment and offers long-term survival for patients with synchronous or metachronous colorectal liver metastases (CLM) [102]. However, only a small percentage (20%) of newly diagnosed patients are suitable candidates for surgery. For those patients who are not eligible for surgery, ablative therapies have emerged as a viable treatment option with acceptable safety and efficacy profiles [103–105]. Despite these advancements, recurrence after CLM treatment remains a significant issue, with a high overall risk of local or distant tumour development following surgical resection or ablation, where resection showed improved local tumour progression (LTP), liver progression-free survival (LPFS), and overall survival (OS) compared to ablative techniques [106, 107]. These studies have not considered as a factor the ablation margins. A recent meta-analysis suggests that achieving a minimal thermal ablation margin of 10 mm yields optimal outcomes for colorectal liver metastases (CLM). However, it is emphasised that a minimal margin of 5 mm is crucial for effective local tumour control [108]. Early recurrences are associated with a poorer prognosis. The use of chemotherapy is crucial in determining the outcome of patients with resectable or unresectable CLM, and it has the potential to convert up to one-third of initially unresectable patients into candidates for potentially curative treatment [109]. Ensuring an accurate evaluation of the response to chemotherapy is of utmost significance in the context of personalised treatment decision-making, particularly in determining the suitability for surgical intervention or the necessity for administering second-line treatment options [110]. In a study conducted by Wei et al., a deep learning (DL) radiomics model was utilised to predict the response of colorectal liver metastases (CLM) patients to various chemotherapy regimens (CAPEOX, mFOLFOX6, FOLFIRI, or XELIRI) based on contrast-enhanced CT scans [111]. The predictions were made by the response evaluation criteria in solid tumours. The results from the validation cohort showed that the deep learning (DL) model had an area under the curve (AUC) of 0.820 (95% confidence interval (CI): 0.681–0.959). Furthermore, when the DL-based model was combined with the measurement of carcinoembryonic antigen (CEA) serum levels, the AUC increased to 0.830 (95% CI: 0.688–0.973). In metastatic cases, treatment with trastuzumab and lapatinib is beneficial in around 70% of patients [112]. Taghavi et al. developed a machine learning (ML)-based radiomics analysis of pretreatment CT scans combined with patients’ clinical features to predict early local tumour progression after ablation treatment. Within this study, up to 5 nodules per patient with a maximum diameter of 30 mm were included. The authors reported a concordance index of 0.79 (95% CI: 0.78–0.80) in the validation cohort for their predictive model [113]. Anti-epidermal growth factor receptor (EGFR) therapies have demonstrated efficacy in the treatment of colorectal liver metastases (CLM) with wild-type RAS mutational status. However, there remains a significant demand for dependable biomarkers that can accurately assess the risk-benefit ratio of these therapies on an individual patient basis with AI assistance [114]. Dercle et al. developed an AI model using ML techniques. This model was designed to assess changes in tumour phenotype based on interval CT scan images taken at baseline and 8 weeks later. The findings of this study indicate that the developed model successfully predicted sensitivity to anti-EGFR therapy with a sensitivity value of 0.80 and a 95% confidence interval (CI) ranging from 0.69 to 0.94. Additionally, the model demonstrated a statistically significant association with overall survival (p < 0.05), further emphasising its potential clinical relevance in prognostic applications [115]. The artificial neural network (ANN) model created by Spelt et al. was used to retrospectively examine a cohort of 241 patients who had undergone liver resection for colorectal liver metastasis (CLM) in a single-centre setting. Out of the 28 potential risk variables assessed, the ANN model identified 6 variables (age, preoperative chemotherapy, size of largest metastasis, haemorrhagic complications, preoperative CEA level, and number of metastases) that were more effective in predicting survival compared to the Cox regression model. The ANN model achieved a C-index of 0.72, indicating improved accuracy in survival prediction, while the Cox regression model achieved a C-index of 0.66 [116].
Future perspectives of AI in liver cancer
Exploring AI’s potential in medicine, future perspectives encompass precision diagnostics, personalised treatments, and transformative healthcare innovations. AI has the potential to greatly aid radiologists in the identification of colorectal liver metastases (CLM) by analysing medical images. In this regard, AI algorithms can play a crucial role in automatically segmenting the liver, which can enable a comprehensive preoperative evaluation of the future liver volume and the resection of liver tissue for the surgical team. This aids in the development of an effective treatment strategy. Additionally, AI-based simulations and educational tools can assist in the training of young healthcare professionals specialising in both surgical and oncological fields. These tools can enhance their ability to accurately diagnose and treat, thereby improving patient outcomes [117–119]. AI-driven therapeutic interventions may prove advantageous for patients who have been deemed unsuitable candidates for surgical treatments. Leveraging computational analysis of genetic and molecular data, AI has the potential to identify plausible pharmaceutical interventions [120]. Recent advancements in the understanding of spatially varying perfusion coefficients, which can be obtained through imaging modalities, have the potential to significantly enhance precision medicine on a global scale. By integrating this knowledge into image analysis techniques, it becomes possible to evaluate the heterogeneity of perfusion in biological tissues with greater accuracy and precision. This breakthrough finding holds immense promise in revolutionising the field of precision medicine and has far-reaching implications for improving diagnostic capabilities and treatment strategies worldwide [121, 122]. Detecting these processes at an early stage could reveal the development of metastatic lesions within the hepatic parenchyma [122, 123]. The advent of robotic systems presents promising avenues for addressing challenges associated with procedural planning, instrument manipulation, precise probe placement, and margin assessment. This technological integration holds the potential to enhance precision and accuracy in liver interventions, thereby contributing to improved overall outcomes in the realm of medical procedures [124].
While existing studies have demonstrated the role of AI and image processing in liver cancer with promising results, it is important to note that many of these studies are not of high-level evidence, such as randomised controlled trials (RCTs). Despite this limitation, there remains a significant need for standardisation and the conduct of large-scale randomised trials to validate and further elucidate the findings from these studies. Standardisation of AI algorithms, imaging protocols, and data analysis methods is crucial to ensure reproducibility and comparability across studies. Additionally, large, randomised trials are essential to assess the clinical utility and effectiveness of AI-based approaches in improving diagnosis, treatment guidance, and prognosis in liver cancer patients. Only through rigorous standardisation and high-quality evidence from large randomised trials can the full potential of AI and image processing in liver cancer be realised.
Conclusions
The integration of AI in liver cancer holds immense promise, offering advancements in early detection, treatment stratification, and prognostic assessment. While preliminary results are encouraging, standardising AI methodologies is crucial. Continued research is essential to validate and refine AI algorithms across diverse patient populations. The integration of AI into routine clinical practice for liver cancer requires comprehensive studies to establish efficacy, reliability, and adherence to standardised protocols. AI promises a more individualised and secure approach to patient care, and collaborative efforts in research and clinical validation will shape the future landscape of AI-driven innovations in liver cancer management.
Acknowledgments
Platon Dimopoulos and Admir Mulita had equal contribution in the preparation of this article.
Funding
No external funding.
Ethical approval
Not applicable.
Conflict of interest
The authors declare no conflict of interest.
References
1. Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2021; 71: 209-49.
2.
World Health Organization. Mortality Database. WHO Statistical Information System. 2008. Available at: http:// www.who.int/whosis.
3.
Bosch FX, Ribes J, Cleries R, et al. Epidemiology of hepatocellular carcinoma. Clin Liver Dis 2005; 9: 191-211.
4.
Thomas London W, Petrick JL, McGlynn KA. Liver cancer. In: Cancer Epidemiology and Prevention. 4th ed. Thun M, Linet MS, Cerhan JR, et al. (eds.). Oxford University Press 2018; 635-60.
5.
Eslam M, Sanyal AJ, George J, et al. MAFLD: a consensus-driven proposed nomenclature for metabolic associated fatty liver disease. Gastroenterology 2020; 158: 1999-2014.
6.
Colombo M, de Franchis R, Del Ninno E, et al. Hepatocellular carcinoma in Italian patients with cirrhosis. N Engl J Med 1991; 325: 675-80.
7.
Brunt E, Aishima S, Clavien PA, et al. cHCC-CCA: consensus terminology for primary liver carcinomas with both hepatocytic and cholangiocytic differentation. Hepatology 2018; 68: 113-26.
8.
Khan SA, Toledano MB, Taylor-Robinson SD. Epidemiology, risk factors, and pathogenesis of cholangiocarcinoma. HPB (Oxford) 2008; 10: 77-82.
9.
Rahnemai-Azar AA, Weisbrod A, Dillhoff M, et al. Intrahepatic cholangiocarcinoma: molecular markers for diagnosis and prognosis. Surg Oncol 2017; 26: 125-37.
10.
Earnest F, Johnson C. Hepatic epithelioid hemangioendothelioma. Radiology 2006; 240: 295-8.
11.
Paolantonio P, Laghi A, Vanzulli A, et al. MRI of hepatic epithelioid hemangioendothelioma. J Magn Reson Imaging 2014; 40: 552-8.
12.
Prabhu VR, Burt AR. Pathology of liver tumors. Surgery 2007; 25: 10-5.
13.
Burt AD, Portmann BC, Ferrell LB. MacSween’s Pathology of the Liver. 5th edn. London 2006; 761-814.
14.
Mark SP, Richard LB, Sheila CR. Hepatic angiosarcoma: findings on multiphasic contrast-enhanced helical CT do not mimic hepatic hemangioma. AJR 2000; 175: 165-70.
15.
Mani H, Van Thiel DH. Mesenchymal tumors of the liver. Clin Liver Dis 2001; 5: 219-57.
16.
Makhlouf HR, Abdul-Al HM, Wang G, Goodman ZD. Calcifying nested stromal-epithelial tumors of the liver: a clinicopathologic, immunohistochemical, and molecular genetic study of 9 cases with a long-term follow-up. Am J Surg Pathol 2009; 33: 976-83.
17.
Caplin ME, Buscombe JR, Hilson AJ, et al. Carcinoid tumour. Lancet 1998; 352: 799-805.
18.
Clark AM, Ma B, Taylor DL, et al. Liver metastases: microenvironments and ex-vivo models. Exp Biol Med (Maywood) 2016; 241: 1639-52.
19.
Chaffer CL, Weinberg RA. A perspective on cancer cell metastasis. Science 2011; 331: 1559-64.
20.
Bosch FX, Ribes J, Diaz M, Cleries R. Primary liver cancer: worldwide incidence and trends. Gastroenterology 2004; 127 (5 Suppl 1); S5-16.
21.
Hallet J, How Lim Law C, Cukier M, et al. Exploring the rising incidence of neuroendocrine tumors: a population‐based analysis of epidemiology, metastatic presentation, and outcomes. Cancer 2015; 121: 589-97.
22.
Gupta GP, Massague J. Cancer metastasis: building a framework. Cell 2006; 127: 679-95.
23.
McMillan D, McArdle C. Epidemiology of colorectal liver metastases. Surg Oncol 2007; 16: 3-5.
24.
Hendriks L, Derks J, Postmus P, et al. Single organ metastatic disease and local disease status, prognostic factors for overall survival in stage IV non-small cell lung cancer: results from a population-based study. Eur J Cancer 2015; 51: 2534-44.
25.
Kulaylat AN, Schubart JR, Stokes AL, et al. Overall survival by pattern of recurrence following curative intent surgery for colorectal liver metastasis. J Surg Oncol 2014; 110: 1011-5.
26.
Manfredi S, Lepage C, Hatem C, et al. Epidemiology and management of liver metastases from colorectal cancer. Ann Surg 2006; 244: 254-9.
27.
Aldrighetti L, Castoldi R, DiPalo S, et al. Prognostic factors for long-term outcome of hepatic resection for colorectal liver metastases. Chir Ital 2005; 57: 555-70.
28.
Wyld L, Gutteridge E, Pinder SE, et al. Prognostic factors for patients with hepatic metastases from breast cancer. Br J Cancer 2003; 89: 284-90.
29.
Goodfellow I, Bengio Y, Courville A. Deep Learning. The MIT Press, Cambridge, USA, 2016.
30.
Wang F, Preininger A. AI in health: state of the art, challenges, and future directions. Yearb Med Inform 2019; 28: 16-26.
31.
Lee P, Bubeck S, Petro J. Benefits, limits, and risks of GPT-4 as an AI chatbot for medicine. N Engl J Med 2023; 388: 1233-39.
32.
Murdoch TB, Detsky AS. The inevitable application of big data to health care. JAMA 2013; 309: 1351-2.
33.
Jiang F, Jiang Y, Zhi H, et al. Artificial intelligence in healthcare: past, present and future. Stroke Vasc Neurol 2017; 21: 230-43.
34.
Johnson KW, Soto JT, Glicksberg BS, et al. Artificial intelligence in cardiology. J Am Coll Cardiol 2018; 71: 2668-79.
35.
Zhao X, Liang P, Yong L, et al. Radiomics study for differentiating focal hepatic lesions based on unenhanced CT images. Front Oncol 2022; 12: 650797.
36.
Oyama A, Hiraoka Y, Obayashi I, et al. Hepatic tumor classification using texture and topology analysis of non-contrast-enhanced three-dimensional T1-weighted MR images with a radiomics approach. Sci Rep 2019; 9: 8764.
37.
Nie P, Yang G, Guo J, et al. A CT-based radiomics nomogram for differentiation of focal nodular hyperplasia from hepatocellular carcinoma in the non-cirrhotic liver. Cancer Imaging 2020; 20: 20.
38.
Ding Z, Lin K, Fu J, et al. An MR-based radiomics model for differentiation between hepatocellular carcinoma and focal nodular hyperplasia in non-cirrhotic liver. World J Surg Oncol 2021; 19: 181.
39.
Cannella R, Borhani AA, Minervini MI, et al. Evaluation of texture analysis for the differential diagnosis of focal nodular hyperplasia from hepatocellular adenoma on contrast-enhanced CT images. Abdominal Radiol 2019; 44: 1323-30.
40.
Cannella R, Rangaswamy B, Minervini MI, et al. Value of texture analysis on gadoxetic acid–enhanced MRI for differentiating hepatocellular adenoma from focal nodular hyperplasia. Am J Roentgenol 2019; 212: 538-46.
41.
Wang X, Wang S, Yin X, Zheng Y. MRI-based radiomics distinguish different pathological types of hepatocellular carcinoma. Comput Biol Med 2022; 141: 105058.
42.
Gatos I, Tsantis S, Karamesini M, et al. Focal liver lesions segmentation and classification in non-enhanced T2-weighted MRI. Med Phys 2017; 44: 3695-705.
43.
Chu H, Liu Z, Liang W, et al. Radiomics using CT images for preoperative prediction of futile resection in intrahepatic cholangiocarcinoma. Eur Radiol 2021; 31: 2368-76.
44.
Qin H, Hu X, Zhang J, et al. Machine-learning radiomics to predict early recurrence in perihilar cholangiocarcinoma after curative resection. Liver Int 2021; 41: 837-50.
45.
Hao X, Liu B, Hu X, et al. A radiomics-based approach for predicting early recurrence in intrahepatic cholangiocarcinoma after surgical resection: a multicenter study. Annu Int Conf IEEE Eng Med Biol Soc 2021; 2021: 3659-62.
46.
Bodard S, Liu Y, Guinebert S, et al. Performance of radiomics in microvascular invasion risk stratification and prognostic assessment in hepatocellular carcinoma: a meta-analysis. Cancers (Basel) 2023; 15: 743.
47.
Zhang D, Zhang XY, Duan YY, et al. An overview of ultrasound-derived radiomics and deep learning in liver. Med Ultrason 2023; 25: 445-52.
48.
Gillies RJ, Schabath MB. Radiomics improves cancer screening and early detection. Cancer Epidemiol Biomarkers Prev 2020; 29: 2556-67.
49.
Hu W, Yang H, Xu H, Mao Y. Radiomics based on artificial intelligence in liver diseases: where we are? Gastroenterol Rep (Oxf) 2020; 8: 90-7.
50.
Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol 2017; 14: 749-62.
51.
Castaldo A, De Lucia DR, Pontillo G, et al. State of the art in artificial intelligence and radiomics in hepatocellular carcinoma. Diagnostics (Basel) 2021; 11: 1194.
52.
Noguchi T, Higa D, Asada T, et al. Artificial intelligence using neural network architecture for radiology (AINNAR): classification of MR imaging sequences. Jpn J Radiol 2018; 36: 691-7.
53.
Nakaura T, Higaki T, Awai K, et al. A primer for understanding radiology articles about machine learning and deep learning. Diagn Interv Imaging 2020; 101: 765-70.
54.
Chaudhury S, Krishna AN, Gupta S, et al. Effective image processing and segmentation-based machine learning tech- niques for diagnosis of breast cancer. Comput Math Methods Med 2022; 2022: 6841334.
55.
Chlebus G, Schenk A, Moltz JH, et al. Automatic liver tumor segmentation in CT with fully convolutional neural networks and object-based postprocessing. Sci Rep 2018; 8: 15497.
56.
Afza F, Sharif M, Khan MA, et al. Multiclass skin lesion classification using hybrid deep features selection and extreme learning machine. Sensors 2022; 22: 799.
57.
Khan MA, Muhammad K, Sharif M, et al. Intelligent fusion-assisted skin lesion localization and classification for smart healthcare. Neural Comput Appl 2021; 36: 37-52.
58.
Attique Khan M, Sharif M, Akram T, et al. A two-stream deep neural network-based intelligent system for complex skin cancer types classification. Int J Intell Syst 2021. https://doi.org/10.1002/int.22691.
59.
Syed HH, Khan MA, Tariq U, et al. A rapid artificial intelligence-based computer-aided diagnosis system for COVID-19 classification from CT images. Behav Neurol 2021; 2021: 2560388.
60.
Sakthi Saravanan B, Meenakshi R. OPBS-SSHC: outline preservation based segmentation and search based hybrid classification techniques for liver tumor detection. Multimedia Tools Appl 2020; 79: 22497-523.
61.
Naeem S, Ali A, Qadri S, et al. Machine-learning based hybrid-feature analysis for liver cancer classification using fused (MR and CT) images. Appl Sci-Basel 2020; 10: 3134.
62.
Lin YS, Huang PH, Chen YY. Deep learning-based hepatocellular carcinoma histopathology image classification: accuracy versus training dataset size. IEEE Access 2021; 9: 33144-57.
63.
Lu LQ, Daigle BJ. Prognostic analysis of histopathological images using pre-trained convolutional neural networks: application to hepatocellular carcinoma. PeerJ 2020; 8: e8668.
64.
Lin HX, Wei C, Wang GX, et al. Automated classification of hepatocellular carcinoma differentiation using multiphoton microscopy and deep learning. J Biophotonics 2019; 12: e201800435.
65.
da Nóbrega RV, Rebouças Filho PP, Rodrigues MB, et al. Lung nodule malignancy classification in chest computed tomography images using transfer learning and convolutional neural networks. Neural Comput Appl 2020; 32: 11065-82.
66.
Zhang HQ, Han L, Chen K, et al. Diagnostic efficiency of the breast ultrasound computer-aided prediction model based on convolutional neural network in breast cancer. J Digit Imaging 2020; 33: 1218-23.
67.
Gazioglu BSA, Kamasak ME. Effects of objects and image quality on melanoma classification using deep neural networks. Biomed Signal Process Control 2021; 67: 102530.
68.
Takeuchi M, Seto T, Hashimoto M, et al. Performance of a deep learning-based identification system for esophageal cancer from CT images. Esophagus 2021; 18: 612-20.
69.
Niu J, Li H, Zhang C, Li D. Multi-scale attention-based convolutional neural network for classification of breast masses in mammograms. Med Phys 2021; 48: 3878-92.
70.
Shahidi F, Mohd Daud S, Abas H, et al. Breast cancer classification using deep learning approaches and histopathology image: a comparison study. IEEE Access 2020; 8: 187531-52.
71.
Li X, Shen X, Zhou Y, et al. Classification of breast cancer histopathological images using interleaved DenseNet with SENet (IDSNet). PLoS One 2020; 15: e0232127.
72.
Cui Z, Leng J, Liu Y, et al. SKNet: detecting rotated ships as key points in optical remote sensing images. IEEE Trans Geosci Remote Sens 2021; 59: 8826-40.
73.
Ma Y, Zhang Y. Insulator detection algorithm based on improved Faster-RCNN. J Comput Appl 2022; 42: 631-7.
74.
Okamura Y, Yamamoto Y, Sugiura T, et al. Novel patient risk factors and validation of a difficulty scoring system in laparoscopic repeat hepatectomy. Sci Rep 2019; 9: 17653.
75.
Hu M, Chen K, Zhang X, et al. Robotic, laparoscopic or open hemihepatectomy for giant liver haemangiomas over 10 cm in diameter. BMC Surg 2020; 20: 93.
76.
Filmann N, Walter D, Schadde E, et al. Mortality after liver surgery in Germany. Br J Surg 2019; 106: 1523-9.
77.
Okuda Y, Taura K, Seo S, et al. Usefulness of operative planning based on 3-dimensional CT cholangiography for biliary malignancies. Surgery 2015; 158: 1261-71.
78.
Fang CH, Tao HS, Yang J, et al. Impact of three-dimensional reconstruction technique in the operation planning of centrally located hepatocellular carcinoma. J Am Coll Surg 2014; 220: 28-37.
79.
Hallet J, Soler L, Diana M, et al. Trans-thoracic minimally invasive liver resection guided by augmented reality. J Am Coll Surg 2015; 220: e55-60.
80.
Phutane P, Buc E, Poirot K, et al. Preliminary trial of augmented reality performed on a laparoscopic left hepatectomy. Surg Endosc 2017; 32: 514-5.
81.
Banz VM, Müller PC, Tinguely P, et al. Intraoperative image-guided navigation system: development and applicability in 65 patients undergoing liver surgery. Langenbeck’s Arch Surg 2016; 401: 495-502.
82.
Pessaux P, Diana M, Soler L, et al. Towards cybernetic surgery: robotic and augmented reality-assisted liver segmentectomy. Langenbeck’s Arch Surg 2014; 400: 381-5.
83.
Meijerink MR, Puijk RS, van Tilborg AAJM, et al. Radiofrequency and microwave ablation compared to systemic chemotherapy and to partial hepatectomy in the treatment of colorectal liver metastases: a systematic review and meta-analysis. Cardiovasc Intervent Radiol 2018; 41: 1189-204.
84.
Tanis E, Nordlinger B, Mauer M, et al. Local recurrence rates after radiofrequency ablation or resection of colorectal liver metastases. Analysis of the European Organisation for Research and Treatment of Cancer #40004 and #40983. Eur J Cancer 2014; 50: 912-9.
85.
Wen R, Chui CK, Ong SH, et al. Projection-based visual guidance for robot-aided RF needle insertion. Int J Comput Assist Radiol Surg 2013; 8: 1015-25.
86.
Wen R, Tay WL, Nguyen BP, et al. Hand gesture guided robot-assisted surgery based on a direct augmented reality interface. Comput Methods Programs Biomed 2014; 116: 68-80.
87.
Sarti M, Brehmer WP, Gay SB. Low-dose techniques in CT-guided interventions. Radiographics 2012; 32: 1109-19.
88.
de Baere T, Tselikas L, Catena V, et al. Percutaneous thermal ablation of primary lung cancer. Diagn Interv Imaging 2016; 97: 1019-24.
89.
Pott PP, Scharf HP, Schwarz MLR. Today’s state of the art in surgical robotics. Comput Aided Surg 2005; 10: 101-32.
90.
Cleary K, Melzer A, Watson V, et al. Interventional robotic systems: applications and technology state-of-the-art. Minim Invasive Ther Allied Technol 2006; 15: 101-13.
91.
Cornelis F, Takaki H, Laskhmanan M, et al. Comparison of CT Fluoroscopy-guided manual and CT-guided Robotic positioning system for in vivo needle placements in swine liver. Cardiovasc Intervent Radiol 2015; 38: 1252-60.
92.
Beyer LP, Wiggermann P. Planning and guidance: new tools to enhance the human skills in Interventional oncology. Diagn Interv Imaging 2017; 98: 583-8.
93.
Bodard S, Guinebert S, Petre E, et al. Percutaneous liver interventions with robotic systems: a systematic review of available clinical solutions. Br J Radiol 2023; 96: 20230620.
94.
Global Burden of Disease Liver Cancer Collaboration. Akinyemiju T, Abera S, Ahmed M, et al. The burden of primary liver cancer and underlying etiologies from 1990 to 2015 at the global, regional, and national level: results from the global burden of disease study 2015. JAMA Oncol 2017; 3: 1683-91.
95.
Lee MW, Rhim H, Cha DI, et al. Percutaneous radiofrequency ablation of hepatocellular carcinoma: fusion imaging guidance for management of lesions with poor conspicuity at conventional sonography. AJR Am J Roentgenol 2012; 198: 1438-44.
96.
Yu NC, Lu DSK, Raman SS, et al.. Hepatocellular carcinoma: Microwave ablation with multiple straight and loop antenna clusters--pilot comparison with Pathologic findings. Radiology 2006; 239: 269-75.
97.
Cornelis FH, Solomon SB. Image guidance in interventional radiology: back to the future? Diagn Interv Imaging 2020; 101: 429-30.
98.
Minier C, Hermida M, Allimant C, et al. Software-based assessment of tumor margins after percutaneous thermal ablation of liver tumors: a systematic review. Diagn Interv Imaging 2022; 103: 240-50.
99.
Mehta N, Dodge JL, Hirose R, et al. Predictors of low risk for dropout from the liver transplant waiting list for hepatocellular carcinoma in long wait time regions: implications for organ allocation. Am J Transplant 2019; 19: 2210-8.
100.
Feng S, Goodrich NP, Bragg-Gresham JL, et al. Characteristics associated with liver graft failure: the concept of a donor risk index. Am J Transplant 2006; 6: 783-90.
101.
Pérez-Ortiz M, Cruz-Ramírez M, Ayllón-Terán MD, et al. An organ allocation system for liver transplantation based on ordinal regression. Appl Soft Comput 2014; 14: 88-98.
102.
Abdalla EK, Vauthey JN, Ellis LM, et al. Recurrence and outcomes following hepatic resection, radiofrequency ablation, and combined resection/ablation for colorectal liver metastases. Ann Surg 2004; 239: 818-25.
103.
Alberts SR, Poston GJ. Treatment advances in liver-limited metastatic colorectal cancer. Clin Colorect Cancer 2011; 10: 258-65.
104.
Tomlinson JS, Jarnagin WR, DeMatteo RP, et al. Actual 10-year survival after resection of colorectal liver metastases defines cure. J Clin Oncol 2007; 25: 4575-80.
105.
Di Martino M, Rompianesi G, Mora-Guzmán I, et al. Systematic review and meta-analysis of local ablative therapies for resectable colorectal liver metastases. Eur J Surg Oncol 2020; 46: 772-81.
106.
Sorbye H. Recurrence patterns after resection of liver metastases from colorectal cancer. Recent Results Cancer Res 2014; 203: 243-52.
107.
Bredt LC, Rachid AF. Predictors of recurrence after a first hepatectomy for colorectal cancer liver metastases: a retrospective analysis. World J Surg Oncol 2014; 12: 391.
108.
Chlorogiannis DD, Sotirchos VS, Georgiades C, et al. The importance of optimal thermal ablation margins in colorectal liver metastases: a systematic review and meta-analysis of 21 studies. Cancers (Basel) 2023; 15: 5806.
109.
Folprecht G, Gruenberger T, Bechstein W, et al. Survival of patients with initially unresectable colorectal liver metastases treated with FOLFOX/cetuximab or FOLFIRI/cetuximab in a multidisciplinary concept (CELIM study). Ann Oncol 2014; 25: 1018-25.
110.
Bonanni L, de Liguori Carino N, Deshpande R, et al. A comparison of diagnostic imaging modalities for colorectal liver metastases. Eur J Surg Oncol 2014; 40: 545-50.
111.
Wei J, Cheng J, Gu D, et al. Deep learning-based radiomics predicts response to chemotherapy in colorectal liver metastases. Med Phys 2021; 48: 513-22.
112.
Meric-Bernstam F, Hurwitz H, Raghav KPS, et al. Pertuzumab plus trastuzumab for HER2-amplified metastatic colorectal cancer (MyPathway): an updated report from a multicentre, open-label, phase 2a, multiple basket study. Lancet Oncol 2019; 20: 518-30.
113.
Taghavi M, Staal F, Gomez Munoz F, et al. CT-based radiomics analysis before thermal ablation to predict local tumor progression for colorectal liver metastases. Cardiovasc Intervent Radiol 2021; 44: 913-20.
114.
Martinelli E, Ciardiello D, Martini G, et al. Implementing anti-epidermal growth factor receptor (EGFR) therapy in metastatic colorectal cancer: challenges and future perspectives. Ann Oncol 2020; 31: 30-40.
115.
Dercle L, Lu L, Schwartz LH, et al. Radiomics response signature for identification of metastatic colorectal cancer sensitive to therapies targeting EGFR pathway. J Natl Cancer Inst 2020; 112: 902-12.
116.
Spelt L, Nilsson J, Andersson R, Andersson B. Artificial neural networks - a method for prediction of survival following liver resection for colorectal cancer metastases. Eur J Surg Oncol 2013; 39: 648-54.
117.
Laino ME, Fiz F, Morandini P, et al. A virtual biopsy of liver parenchyma to predict the outcome of liver resection. Updates Surg 2023; 75: 1519-31.
118.
Brunese MC, Fantozzi MR, Fusco R, et al. Update on the applications of radiomics in diagnosis, staging, and recurrence of intrahepatic cholangiocarcinoma. Diagnostics 2023; 13: 1488.
119.
Li Y, Gong J, Shen X, et al. Assessment of primary colorectal cancer CT radiomics to predict metachronous liver metastasis. Front Oncol 2022; 12: 861892.
120.
Singh M, Dalal M, Sodhi GS. Estimation of clinical size of breast tumour lesions using contrast enhanced magnetic resonance imaging: delineation of tumor boundaries. In UMBC Student Collection; UMBC: Baltimore, MD, USA, 2021.
121.
Lugano R, Ramachandran M, Dimberg A. Tumor angiogenesis: causes, consequences, challenges, opportunities. Cell Mol Life Sci 2020; 77: 1745-70.
122.
Zhao Y, Adjei AA. Targeting angiogenesis in cancer therapy: moving beyond vascular endothelial growth factor. Oncologist 2015; 20: 660-73.
123.
Fleischer JR, Schmitt AM, Haas G, et al. Molecular differences of angiogenic versus vessel co-opting colorectal cancer liver metastases at single-cell resolution. Mol Cancer 2023; 22: 17.
124.
Stewart C, Wong P, Warner S, et al. Robotic minor hepatectomy: optimizing outcomes and cost of care. HPB (Oxford) 2021; 23: 700-6.
