Abstract
The usage of expert systems to aid in medical decisions has been employed since 1980s in distinct applications. With the high demands of medical care and limited human resources, these technologies are required more than ever. Skin cancer has been one of the pathologies with higher growth, which suffers from lack of dermatology experts in most of the affected geographical areas. A permanent record of examination that can be further analyzed are medical imaging modalities. Most of these modalities were also assessed along with machine learning classification methods. It is the aim of this research to provide background information about skin cancer types, medical imaging modalities, data mining and machine learning methods, and their application on skin cancer imaging, as well as the disclosure of a proposal of a multi-imaging modality decision support system for skin cancer diagnosis and treatment assessment based in the most recent available technology. This is expected to be a reference for further implementation of imaging-based clinical support systems.
TopBackground
Like any other cell in the human body, skin cells are subjected to different types of mechanisms that regulate their development and replacement, if considered needed.
Typically, physiological pathways are triggered to induce apoptosis of malfunctioning cells and destroy it before causing harm. If these defense mechanisms fail, the defective cell can grow out of control and generate a skin neoplasm. (Hunter et al., 2002a)
An abnormal cell grow is not inevitably indicative of the appearance of a cancerous tissue. Benign or malignant tumors can arise, and its differentiation should be clarified. Malignant masses are generally referred to as cancers and are histologically characterized by cells considerably dissimilar from the ones of its mother-tissue. There is a clear tendency to grow and multiply at an excessive rate, as well as to eventually infiltrate neighbor tissues and close-by vascular and lymphatic structures. This feature, i.e., metastization, allows it to spread to different organs and originate a new tumor focus, complicating its treatment and increasing the associated life-risk. Contrarily, benign neoplastic cells tend to be somewhat like its tissue of origin, presenting decent cell differentiation. Their growth rate is considered slower when compared to malignant ones, and the ability to metastasize is absent. In fact, these tumor types have a tendency towards local expansion, pushing adjacent structures. Thus, normally, it does not represent a threat for its host. (Crowley, 2013; L Kemp et al., 2015)
Anatomical changes are guaranteed to happen with the emergence of a tumorous mass. Still, the occurrence of physiological shifts is far more relevant, especially if in the presence of a cancerous lesions. (Baba & Catoi, 2007a) Contrarily to what most could hypothesize, the neoplastic mechanisms are not uniquely controlled by genes that underwent mutations. Genes that retained its “normal” structure can have a detrimental influence, as the promotion of the expression of proteins, normally included in regular cellular processes, but at inappropriate occasions and with the inappropriate extent. Thus, enhancing carcinogenesis. (Moasser, 2014)
Key Terms in this Chapter
Expert Systems: Computational system build to mimic the capability of a human specialist to perform a decision.
Datawarehouse: Electronic system for the collection and manipulation of large amounts of data.
Accuracy: Parameter that represents the number of individuals correctly diagnosed among a given group, by a clinician or decision support system.
Merkel Cells: Epidermis skin cells that play a role in human sensing and work as mechanoreceptors.
Sensitivity: Measure of the ability of a clinician or decision support system to correctly identify those who have the disease.
Neoplasm: Atypical growth and agglomeration of cells.
Decision Support System: Computational system employed to assist, sustain decisions, and select the best course of action.
Melanocytes: Epidermis skin cells that produce and store melanin, a pigment that absorbs ultraviolet rays, blocking its nefarious actions.
Supervised Learning: A machine learning algorithm that learns from labeled data included in a training set.
Basal Cells: Epidermis skin cells responsible for producing new skin cells.
User Interface: Platform created for the interaction of a human operator with a computer.
Metastization: The ability of malignant cells to infiltrate neighbor tissues and close-by vascular and lymphatic structures to spread to distant organs.
Langerhans Cells: Epidermis skin cells with immunological functions.
Datamart: Subset of a data warehouse focused on a specific topic of information, per example, patient information.
Echogenicity: The capability of a given tissue to bounce an echo. This ability is low for hypoechoic structures (denser) and high for hyperechoic structures.
Dysplasia: Abnormal cell development, normally associated to the appearance of pre-cancerous or cancerous lesions.
Data Mining: Process involved on the harvesting of information from unprocessed data.
Specificity: Measure of the ability of a clinician or decision support system to correctly identify those who do not have the disease.
Machine Learning: Scientific area of Artificial Intelligence focused on the development of algorithms and/or systems with the ability to learn, modify in an automatic manner and reach decisions, based on data patterns.
Unsupervised Learning: A machine learning algorithm that learns from unlabeled data, learning and detecting characteristic data patterns on its own.
Business Intelligence: Software tools that can analyze large data quantities and retrieve information to reach knowledgeable conclusions.
OLAP Cube: Array with large amounts of data used for multidimensional analysis.
Fluorophores: Fluorescence compound that can radiate light upon excitation with a light source.
Inference Engine: Section of a machine learning system destined to logically interpret a given data set and retrieve information from it.
Computer-Aided Diagnosis: System that supports medical diagnosis and helps in treatment selection.
Keratinocytes: Epidermis skin cells that originate from the differentiation of basal cells, composing approximately 85% of epidermis. It is the skin constituent responsible for the synthesis of vitamin D and keratin production.
Deep Learning: Subfield of machine learning that mimics the workings of the human brain in process of data analysis, interpretation and decision making.