黑料网

Machine Learning (ML); Nano-medicine; Cancer; Deep Learning (DL); Artificial Intelligence (AI); Liver

The liver is the sixth most prevalent location for primary cancer and there are around 800,000 cases identified worldwide each year. Liver cancer is the fourth greatest cause of cancer related death worldwide. Hepatocellular Carcinoma (HCC), Cholangio Carcinoma (CCA) and Metastatic Liver cancer (LM) are three general categories for liver malignancies. HCC and CCA make up 80%-90% and 10%-15%, respectively, of these initial liver malignancies [1]. The American Association for the Study of Liver Diseases (AASLD) currently advises surveillance of HCC based on high serum fetoprotein levels (>20 ng/ML) and liver nodule (>1 cm) detection through abdominal ultrasonography in liver disease patients. To date, HCC remains the most common type of liver cancer. With excellent specificity (96%) and sensitivity (89%) for HCC, multiphase Computed Tomography (CT) or Magnetic Resonance Imaging (MRI) can further characterize liver nodules larger than 1 cm with rim hyper-enhancement, peripheral washout, and delayed central enhancement as characteristic hallmarks. The Liver Reporting and Data System (LI-RADS) score that offer a diagnostic likelihood of HCC can be further interpreted from this information.

Despite these developments, liver biopsy is still necessary in individuals who do not have cirrhosis since imaging cannot reliably confirm the diagnosis of HCC in these people. Once identified, HCC care relies on a number of variables, including the tumor stage, the patient's performance level, and the fraction of the liver that is functioning, necessitating a customized strategy [2]. Local ablation, resection, and liver transplantation are the current treatment choices for early stage HCC, while systemic treatments and trans-arterial radio embolization (TACE) continue to be the mainstays for advanced illness. The patient's candidature for a liver transplant can also be maintained by using TACE as a bridging medication.

CCA, a very fatal epithelial tumor that develops along the biliary system or inside the hepatic parenchyma and has characteristics of cholangiocyte differentiation, is another primary liver cancer. CCAs can be anatomically classified as intrahepatic (iCCA), perihilar (pCCA), or distal (dCCA). ICCA is proximally localized to the secondorder bile ducts in the liver parenchyma, pCCA can be further anatomically classified by the bismuth Corlette classification and dCCA is localised at the common bile duct below the cystic duct insertion.

Unfortunately, symptoms typically appear at an advanced stage in CCA patients and they are frequently asymptomatic in the early stages of the disease. The main blood biomarker for iCCA diagnosis is Carbohydrate Antigen 19-9 (CA19-9), with values >1000 U/ML indicating advanced illness [3]. Contrast Enhanced Ultrasonography (CEUS), CT, MRI, and conventional Ultrasonography (US) are imaging modalities that can help in the diagnosis of iCCA. The diagnosis of pCCA and dCCA can be made with up to 85% sensitivity using Magnetic Resonance Cholangio Pancreatography (MRCP). In addition to these, biliary cytology or biopsy can be used to confirm CCA; however, because this method only detects pCCA with a sensitivity of 43%, doctors typically combine brush cytology with fine needle aspiration and biopsy to improve diagnostic yield.

People with proven CCA get chemotherapy, surgical excision, or both. Well-chosen people with pCCA may be candidates for liver transplantation. Hepatectomy with regional lymphadenectomy has been tried for iCCA and dCCA, however pancreaticoduodenectomy would be necessary in a small number of carefully chosen patients for pCCA. First line chemotherapy with gemcitabine and cisplatin is given to individuals with advanced illness who are ineligible for the aforementioned treatments. Despite recent advancements in the diagnosis and treatment of CCA, several important clinical and scientific problems remain unresolved [4]. For the detection of this heaptobiliary illness in bile and blood, better and more accurate diagnostic techniques are required. Since there is no proven one size fits all treatment and few options for CCA management, more advanced technology is urgently needed to prevent and treat this deadly disease. Additionally, during treatment, surgical resection and liver transplantation may potentially affect many other organs with serious complications, while chemotherapy patients frequently require molecular profiling as part of their treatment plan.

Furthermore, the liver is a desirable location for metastases because of the abundant portal blood supply. The most frequent liver metastases are Colorectal Cancer (CRC), followed by pancreatic cancer, breast cancer, melanoma and lung cancer. Liver metastases are tumors that have migrated from the organ of origin to the liver. With left sided tumors being more prone to liver metastasis than right sided tumors, CRC is the most frequent cause in individuals with advanced stage cancer [5]. Although the cause of heterogeneity in metastases is not fully understood, it has been suggested that liver metastasis is caused by tumor cells adhering to the target organ through intricate interactions and genetic mutations with resident liver cells for the tumor's engraftment, survival, and progression in the liver. This is accomplished through the following four major phases: i) Microvascular phase for tumour cell intravasation and arrest in sinusoidal vessels, ii) Extravascular pre-angiogenic phase, iii) Angiogenic phase, iv) Proliferation phase allowing tumour cells to metastasize and establish clinically detectable cancer cells.

If patients with liver metastases have severe illness or don't receive treatment, their quality of life will deteriorate and they have a bad prognosis with an almost zero-5 years survival rate. This is due to the fact that liver metastases impede liver function, which in turn causes jaundice, coagulopathy and/or ascites. Depending on the degree of the metastasis and the underlying tumor's origin, the care of liver metastases typically requires more specialised and customised techniques, such as liver resection, ablation, locoregional medicines, chemotherapy and/or targeted therapies. However, the original tumour is mostly responsible for the best disease management, thus accurate diagnosis and identification of the underlying tumour are essential. According to American Society of Clinical Oncology (ASCO) guidelines published in 2022, the Chemotherapy Regimen for Colorectal Cancer Liver Metastasis (CRCLM) includes 5 fluorouracil, irinotecan, or oxaliplatin. In some cases, these have been combined with immunotherapeutic drugs bevacizumab, cetuximab, or panitumumab and patients have shown improved survival compared to those receiving standard therapy alone. Unfortunately, despite these improvements in the diagnosis and treatment of CRCLM, it is critically necessary to target early identification and personalised disease management methods in future directions. We are at the beginning of a revolution in the detection and treatment of cancer thanks to the growing development and use of Artificial Intelligence (AI) in medicine. More precise diagnostics are required to satisfy the needs for personalised therapy, which may be beyond the capabilities of humans, as a result of rising liver cancer heterogeneity and the availability of a greater variety of therapeutic alternatives. Due to its ability to enhance the utilisation of vast, multi parametric data to extract important information and accomplish individualised clinical diagnosis and management decisions for patients, Artificial Intelligence (AI) offers the biggest benefit in this regard.

With the use of AI, significant advancements have been made in the fields of gastroenterology and hepatology to date. AI has been investigated for endoscopic lesions and GI bleeding detection, to identify patients with liver fibrosis, to forecast therapy response, etc. With these developments, AI offers hope for closing the present gaps in the detection and treatment of liver cancer. The use of Machine Learning (ML) and Deep Learning (DL) to the detection and treatment of liver cancer will be the main emphasis of this publication. We will also talk about existing difficulties with AI use in HCC, CCA, and CRCLM as well as potential solutions in the future.

Introduction of artificial intelligence

The definition of Artificial Intelligence (AI) refers to a broad field that includes the use of computer programmes to carry out complicated tasks that may be related to human learning behaviours, uncertainty quantification, probabilistic reasoning, problem solving techniques and knowledge representation (Figure 1).

Figure 1: AI classification and commonly used techniques.

ML is a branch of computational techniques that makes use of datasets to comprehend and develop techniques that learn certain jobs, including categorizing or assessing an outcome. With the use of computing, ML makes it possible to fit prediction models to data to roughly or consciously identify trends [6]. ML may also be broken down into supervised, unsupervised, and reinforcement learning categories. In supervised machine learning, algorithms are trained to fit a model for precise result prediction using labeled data. This can be accomplished using a regression model to forecast a continuous value that is not discrete (for example, fetoprotein levels against liver cancer patient age) or a classification model that employs an algorithm to categorise data into a set of discrete categories (for example, "HCC" or "not HCC") (Figure 2).

Figure 2: Commonly used ML and DL algorithms.

Unsupervised learning, on the other hand, makes use of unlabeled data to find obscure or hidden patterns. Three primary tasks are carried out using unsupervised models: Unlabeled data may be clustered based on input similarities or differences, data can be associated to establish a connection between variables and big data processing can be managed via dimensionality reduction.

The next stage of unsupervised learning is called reinforcement learning and it is characterised by the ability of the learning model to predict the result of a series of decisions through continuous trials and mistakes. Additionally, as the data output is frequently not optimal, loss or cost functions must be used to calculate the prediction and the difference between the prediction and the observation. Testing, validation and training are all part of the process overall. To automatically update algorithm parameters for least loss or cost function and best performance, a model is initially trained using a training dataset (Figure 3).

Figure 3: Flowchart showing the overall process of developing an AI model.

Contrarily, DL is a subclass of ML models that are created using neural networks (NNs, also known as Artificial Neural Networks (ANNs) and Feed Forward Neural Networks), which are made up of mathematical methods that are modeled after the neuronal behaviour in the human brain. A deep neural network is made up of thousands to billions of densely connected computing neurons that are arranged in successive layers. These layers receive data and transmit it to succeeding/hidden layers where it is added, subtracted, divided, or multiplied thousands of times before reaching the output layer, which represents the algorithm's predicted outcome.

Convolutional, pooling and fully linked layers make up CNN's fundamental architecture. These are mostly employed in feature extraction tasks that guarantee translation invariance and retain the spatial layout of the data. It is most frequently employed in image analysis where the features are object forms, such as when automatically identifying localised liver lesions in CT or MRI images.

Artificial intelligence in the detection and treatment of HCC

The extensive range of HCC heterogeneity, which differs from patient to patient in terms of risk factors and a etiology, diversifies and limits how prognosis, treatment, and prediction are carried out. There are currently uncommon prospects in improving HCC that have been investigated, including better diagnostics, better disease prediction, and better disease management in individuals with liver disease to assess the chance of developing HCC.

The ability to diagnose HCC by CT, MRI, and histopathology scans more quickly and with less risk of unfavorable clinical consequences is made possible by advances in computer-aided automation.

Artificial intelligence in the detection and treatment of cholangiocarcinoma

Knowledge, diagnosis, and treatment of CCA have not much improved in the last ten years, and the illness has a very bad prognosis with low 5 years survival rates (7%-20%). Since there are currently no reliable non-invasive diagnostic biomarkers, the only curative alternative is tumour excision, which only qualifies 20%-40% of patients. Although some information has previously been revealed, the key hurdles in CCA care continue to be increasing disease heterogeneity, medication resistance and tumour recurrence postresection; as a result, fresh approaches to characterising these tumours and delivering targeted therapy are necessary. AI modalities might shift the paradigm in illness diagnosis and general care by filling the critical gaps in CCA research [7]. So far, novel biomarkers have been found in serum, bile, saliva and urine samples thanks to the utilisation of omics data.

According to several studies, bodily fluids frequently include circulating nucleic acids due to CCA and apoptotic tumour cells. These circulating nucleic acids may be discovered and used to develop novel diagnostics and tailored treatments that differ from patient to patient.

Cell-free DNA (cfDNA), one of these circulating tiny molecules, has drawn a lot of attention since it has been shown to be directly related to the size and aggressiveness of CCA tumours. As a result, multi-omics data can be further developed and tested using ML algorithms to support clinical research and decision making. Furthermore, these circulating nucleic acids also serve as a platform to identify multiple mutations and subsequently direct clinicians on mutational-based treatment planning in CCA patients.

Artificial intelligence in the diagnosis and treatment of colorectal liver cancer metastases

Despite recent advances in our understanding of CRCLM, many important scientific concerns remain, necessitating further research and attention to ensure accurate early identification and care of this life threatening condition. More than 90% of CRCLM patients are not accurately diagnosed in the early stage due to diagnostic limitations and therefore undergo incomplete endoscopic resections as part of their therapeutic intervention, making surgical resection the only curative option currently available for patients detected with early CRCLM [8]. To get around this, precise prediction algorithms might be created to identify people who are most likely to contract the illness.

There are currently no publications demonstrating the use of ML or DL models from a therapy viewpoint, which further opens up a wealth of prospects for study in this area. AI models have been utilised for faster and more accurate imaging and histological tests in CRCLM patients. One significant use of DL models, however, has been to forecast drug response in a database of cancer cell lines enabling more individualised treatment in clinical cohorts.

The emergence of the 'omics' era has allowed researchers to better understand the most significant oncogenic drivers by using ML and DL algorithms to the widely available databases. Predicting the clinical response to any cancer treatment is frequently a difficulty in cancer patients. Recent research has shown the validity of DL models trained on cell lines to predict medication response using gene expression levels. For a personalised medicine, these DL models are trained to accurately represent the biological connections between various genes.

Current difficulties with using AI to deal with and identifying liver cancer

AI is currently beginning to spread in the field of liver cancer diagnosis and treatment. AI has quickly dominated the world of medical imaging, disease prediction, and management automation in various illnesses at the level of physicians' expertise in their respective disciplines. For instance, in the diagnosis of liver cancer, certain AI algorithms are doing as well as radiologists and pathologists in terms of tumor identification and grading; yet, other AI technologies have not yet been able to effectively navigate the present obstacles before being employed in clinical settings. The lack of open access to many ML and DL models impedes the development of this field. This highlights the requirement for additional open sourced platforms and algorithms for researchers to investigate and employ in order to enhance the study of liver cancer.

The 'black box' characteristic of ML and DL models makes using them for the detection and treatment of liver cancer a significant issue at the moment. High stakes clinical judgments about the diagnosis and treatment planning of liver cancer necessitate decisions that are reliable, impartial, and rational. Despite AI advancements, the majority of ML and DL models lack the interpretability necessary to demonstrate how particular parameters affect the anticipated outcome during the decision-making process.

Numerous tools have been created that leverage fundamental domain principles during analysis to address the problem of interpretability, such as gradient based techniques for image analysis and the SHapley Additive exPlanations model (SHAP values) for data based analysis [9]. Image specific saliency maps have been created during gradient-based analysis to identify certain sections of an image through pixel analysis that lead to the output prediction. This gradient based technique checks to see if computer-based conclusions agree with the domain interpretations of radiologists. In order to overcome the problem of interpretability in data based analysis, SHAP values, which determine the contribution of each variable feature to the anticipated outcome of the model, have been taken into consideration.

Future opportunities: Can AI in nanomedicine lead to a paradigm change to expedite the treatment of liver cancer

Nano-medicines have rapidly gained popularity over the past ten years, giving researchers a fresh push to tackle "undruggable" illnesses faster from the lab to the clinic. With the recent success of COVID-19 vaccines, which went from the lab to the clinic in just 16 weeks, nanotechnology has taken the lead in drug research and is starting to change the accepted standards of care for some infectious and genetic conditions. Despite the significant benefits that can be attributed to nanoparticles in terms of drug delivery, precise targeting, simultaneous administration of several therapeutic agents, and enhancement of pharmaceutical properties, the development process for nanoformulations is extensive, moderately drawn out and still heavily depends on trial and error.

Additionally, the chronic liver disease environment in which liver cancer develops causes present restrictions to preserve medication synergy to vary from patient to patient, which obstructs the path to effective liver cancer nano-medicine development. AI is accelerating optimization and assessment in the field of nano-medicine to overcome these obstacles during the developmental stages.

Additionally, because of their lower dose frequencies, less side effects, focused administration, lower in vivo drug fluctuations, and ability to maintain drug concentration, controlled release nano formulations are rapidly being explored for liver cancer therapy. Despite these nano formulations' benefits, maintaining appropriate drug release kinetics can be challenging because of their complexity. In certain cases, the association between different process factors is also unclear, necessitating time consuming optimisation.

Additionally, nanoparticles have developed as drug carriers to improve cellular uptake in experiments using liver cancer cells. Because most medications are unable to permeate the plasma membrane, the capacity of drug molecules to enter tumor cells is constrained. For a novel formulation to be effective, it must be able to cross cellular barriers, hence AI methods like CNNs have been used to increase cellular absorption in various diseases such triple negative breast, pancreatic, and lung cancers [10]. A good basis for integrating AI with medication development that may be expanded to liver cancer research was established by a recent study that used CNNs to logically choose the optimal nanoparticle based on their cellular absorption analyses.

In conclusion, because to the existing limitations in the diagnosis and treatment of liver cancer caused by the disease heterogeneity, a lack of understanding of the cell of origin, and challenges in the delivery of particular non-toxic drugs to the liver tumour cells, AI may revolutionise the discipline of liver cancer research. Modern AI technologies have made strides in a number of studies on liver cancer, but it would be premature to distinguish fact from fiction until the 'black box' of AI is completely understood and can be used as a standalone treatment in clinical practise. As a result, the ultimate objective of AI is to offer decision support tools that can help doctors understand the complexities of diagnosing and treating liver cancer. Additionally, the above-discussed paradigm shift in the field of AI in nano-medicine may be employed in concert to expedite the discovery and development of the most personalised nano targeted medicines for people with liver cancer.

1. Bakrania A, Joshi N, Zhao X, Zheng G, Bhat M (2023) . Pharmacol Res 189: 106706.

   [] [] []

2. Shinmura R, Matsui O, Kobayashi S, Terayama N, Sanada J, et al. (2005) . Radiology 237: 512-519.

   [] [] []

3. Lee S, Kim YY, Shin J, Hwang SH, Roh YH, et al. (2020) . J Am Coll Radiol 17: 1199-1206.

   [] [] []

4. Manini MA, Sangiovanni A, Martinetti L, Vigano D, La Mura V, et al. (2015) . Liver Transpl 21: 1259-1269.

   [] [] []

5. Nataliya R, Gores GJ (2014) Cholangiocarcinoma. Lancet 383: 2168-2179.
6. Farshidfar F, Zheng S, Gingras MC, Newton Y, Shih J, et al. (2017) . Cell Rep 18: 2780-2794.

   [] [] []

7. Rizvi S, Gores GJ (2013) . Gastroenterology 145: 1215-1229.

   [] [] []

8. Saluja SS, Sharma R, Pal S, Sahni P, Chattopadhyay TK (2007) . Hpb 9: 373-382.

   [] [] []

9. Milette S, Sicklick JK, Lowy AM, Brodt P (2017) . Clin Cancer Res 23: 6390-6399.

   [] [] []

10. van den Eynden GG, Majeed AW, Illemann M, Vermeulen PB, Bird NC, et al. (2013) . Cancer Res 73: 2031-2043.

    [] [] []