Abbreviations: AI: Artificial Intelligence; ML: Machine Learning; DL: Deep Learning; HCC: Hepatocellular Carcinoma; TACE: Transarterial Chemoembolization
Opinion
Dear Editor, due to continuous implementing of medical devices physicians are dealing with tremendous amount of data and clinical information. This is especially true within the oncological setting. Therefore, the management of oncological patients requires that clinical decisions be taken within multidisciplinary teams made up of clinicians, radiologists, geneticists, surgeons, pathologists, psychologists and oncologists. However, some lights may be at the end of the tunnel. Recent development of computer algorithms has reached excellent results and is now able to simulate human cognitive functions, such as learning or problem solving. This processing is called artificial intelligence (AI). AI utilized Machine Learning (ML) and deep learning (DL). The first one, ML is based on the ability of the computer to “learn” and improve from past examples without being programmed. DL is a subset of ML and is computer software that mimics the network of neurons in a brain. In DL, the learning phase occurs through a neural network. For the above reasons is clear that AI is potentially useful for making clinical diagnosis and taking clinical decisions especially in oncology.
We believe that AI could become the new tool for the
management of hepatocellular carcinoma (HCC) helping to predict
the onset, recurrence and prognosis. Recently, Jiménez Pérez M and
Grande RG and their colleagues published a review article showing
how AI could help differentiate between normal liver, chronic liver
disease, cirrhosis and HCC or benign and malignant nodules. AI is
able to in the diagnostic accuracy, tumor staging, treatment planning
by utilizing several types of radiological images (ultrasound, CTscan,
MRI-scan, etc), WHO classifications, histopathological findings
(malignant tumors non-HCC, indeterminate masses, dysplastic
nodules etc.) [1]. Interestingly, the use of AI and ML techniques
has also been applied on the predictivity of response both in
terms of HCC recurrence after resection and after transarterial
chemoembolization (TACE). In the first case, radiomics can improve
predictive accuracy for HCC recurrence after curative resection [2].
Also, the effects of transarterial chemoembolization in patients with
HCC can be predicted by combining clinical data and MR imaging.
The images obtained from CT, MRI or PET exams are converted
into numerical data through radiomics. These data are manipulated with the use of AI for the management of so-called “big data”. With
these techniques is possible to obtain from the integrated analysis
of several radiological imaging the correct indications on which
treatment should be performed to achieve the best clinical response
[3]. Translational research in HCC has introduced amounts of
molecular data. These data come from studies conducted from
patients, tissues, in vitro models (cell lines and organoids) and
in vivo models. In recent years, the diagnostic markers and
therapeutic targets based on genomic mutations, expression of
proteins, genes or metabolites between HCC and healthy tissues
have been proposed. Although this methodology has led to the
discovery of several drugs, that are molecularly targeted, this has
not been the case in HCC. The therapeutic failure of HCC is most
likely the presence of tumor heterogeneity, both among patients
and within them AI could help therefore unveil HCC diagnosis and
correct treatment [4].
Although promising, AI in HCC has some problematics: specific
studies are needed to confirm that the developed algorithm can be
used in clinical practice. Also, multicenter prospective studies are
needed to avoid any bias that may later affect learning. The power
of AI is affected by the fact that data are retrospectively collected
and often that databases are not consistent and homogeneous.
Users need to understand the true usefulness of AI, including its
limitations. In addition, economics and ethical aspects can not be
forgotten. Finally, statistical association does not necessarily mean
a causal link. In conclusion, neural networks are efficient only if the
variables are carefully chosen.
References
- Jiménez Pérez M, Grande RG (2020) Application of artificial intelligence in the diagnosis and treatment of hepatocellular carcinoma: a review. World J Gastroenterol 26(37): 5617-5628.
- Ji GW, Zhu FP, Xu Q, Wang K, Wu MY, et al. (2019) Machine-learning analysis of contrast-enhanced CT radiomics predicts recurrence of hepatocellular carcinoma after resection: A multi-institutional study. EBioMedicine 50(12): 156-165.
- Abajian A, Murali N, Savic LJ, Laage Gaupp FM, Nezami N, et al. (2018) Predicting Treatment Response to Intra-arterial Therapies for Hepatocellular Carcinoma with the Use of Supervised Machine Learning-An Artificial Intelligence Concept. J Vasc Interv Radiol 29(6): 850-857.e1.
- Chen B, Garmire L, Calvisi DF, Chua MS, Kelley RK, et al. (2020) Harnessing big ‘omics’ data and AI for drug discovery in hepatocellular carcinoma. Nat Rev Gastroenterol Hepatol 17(4): 238-251.