AI in Healthcare: A Review (2014-2023)

On the other hands, to illustrate the application of machine learning algorithms for disease diagnosis and prognosis, Esteva et al., (2017) present a case study that focuses on the use of these algorithms in the field of cancer detection. The study utilized a large dataset comprising medical images, electronic health records, and genomic data from patients diagnosed with various types of cancer. The goal was to develop a predictive model that could accurately diagnose cancer and provide prognostic information to aid in treatment planning. Medical images, including X-rays, CT scans, and MRI scans, were collected from multiple healthcare institutions. These images were anonymized and subjected to preprocessing steps, such as image normalization, noise reduction, and image registration, to ensure consistency and improve the quality of the data. Electronic health records, containing information about patient demographics, medical history, laboratory results, and treatment records, were obtained and standardized across the dataset. Genomic data, including gene expression profiles and genetic mutations, were also collected for each patient. For each patient, a set of relevant features was extracted from the medical images, electronic health records, and genomic data. This involved using image processing techniques to extract texture features, shape descriptors, and intensity statistics from the medical images. Text mining and natural language processing techniques were employed to extract relevant information from the electronic health records, such as comorbidities, tumor characteristics, and treatment history. Genomic data were analyzed to identify gene expression patterns and genetic mutations associated with specific types of cancer. Various machine learning algorithms were employed to develop a predictive model for cancer diagnosis and prognosis. These algorithms included support vector machines, random forests, and DL models such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs). The dataset was divided into training, validation, and testing sets to train the models, tune hyperparameters, and evaluate their performance. Model training involved feeding the extracted features and corresponding labels (i.e., cancer diagnosis and prognosis) into the machine learning algorithms. The models learned patterns and relationships within the data to make accurate predictions. Performance metrics such as accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC-ROC) were used to evaluate the models’ performance. Cross-validation techniques, such as k-fold cross-validation, were employed to assess the models’ generalizability and robustness. The developed machine learning models demonstrated promising results in cancer diagnosis and prognosis. The models achieved high accuracy and AUC-ROC values, indicating their ability to accurately distinguish between cancer and non-cancer cases. Furthermore, the models provided prognostic information, such as predicting disease progression, recurrence risk, and survival outcomes. These predictions aided healthcare professionals in designing personalized treatment plans and making informed clinical decisions. The interpretation of the machine learning models involved identifying the most influential features and understanding their contributions to the prediction task. Feature importance analysis, such as permutation importance or feature importance scores from the models, provided insights into the key factors influencing the diagnosis and prognosis. This information could guide further research, validate existing medical knowledge, and uncover novel biomarkers or risk factors associated with specific types of cancer.