Decision Support Systems for Disease Detection and Diagnosis

The last few years have been characterized by a large amount of research activity in the field of healthcare for both the improvement of diagnostic treatments and the development of simple, efficient, and multi-tasking applications. Disease preventive measures, clinical management, health decision-making, prognosis, and treatment should be taken into account for the deployment of a rewarding smart medicine [1].

The combination of biomedical studies and information technology has produced many advantages in treating several pathologies [2,3,4]. The development of accessible, affordable, user-friendly, and accurate solutions secures an entitlement to early screening/treatment, which could improve patient survival rates, especially in the application of computer-aided equipment [5,6]. In particular, decision support systems (DSSs) are becoming important tools in supporting physicians, particularly when both diagnostic signals are complex to analyze (due to their low quality) and a high number of examinations have to be studied, as in the case of mass-screening programs [7,8,9]. DSSs give information to experts “where and when it is needed” by providing models and procedures for supporting practitioners to make a decision that is less prone to errors [10,11]. Therefore, DSSs are expected to assist and support clinicians during the decision-making process by acting as a “second opinion”, combining physicians knowledge with the “knowledge” that is embedded in the system [12,13].

This Special Issue provides a glimpse of the most recent research on the development of computer-aided detection and diagnosis systems able to assist medical staff during clinical routine detection/diagnosis of diseases. Several papers have been submitted, but after a rigorous review process, only eleven papers have met the highly demanding criteria for publication followed by the Editorial Board. The papers selected for inclusion in the Special Issue offer a wide plethora of leading-edge subjects that fall within the topics indicated in the call for papers of the Special Issue.

The paper by Macin et al., “An Accurate Multiple Sclerosis Detection Model Based on Exemplar Multiple Parameters Local Phase Quantization: ExMPLPQ”, proposes a lightweight machine learning system that adopts a novel feature engineering algorithm based on local quantization for multiple sclerosis diagnosis.

To support physicians for neurodegenerative disease detection, an automatic visual system that adopts a low-cost camera is indicated by Cicirelli et al. in “Low-Cost Video-Based System for Neurodegenerative Disease Detection by Mobility Test Analysis”. People are recorded while performing the sit-to-stand test, and the obtained videos are analyzed for the extraction of features based on skeleton joints.

The detection of human posture is the aim of the paper “Human Posture Detection Using Image Augmentation and Hyperparameter-Optimized Transfer Learning Algorithms” by Ogundokun et al. An innovative three-phase model that integrates convolutional neural network transfer learning, image data augmentation, and hyperparameter optimization is conceived for the design of the DSS.

Hussain et al., in the study named “A. Shape-Based Breast Lesion Classification Using Digital Tomosynthesis Images: The Role of Explainable Artificial Intelligence”, have developed and validated a deep learning framework to support medical staff in classifying breast lesions from tomosynthesis images. The conceived DSS takes into account the shape of the lesion and encircles the potential growth pattern of the tumorous regions on the image under test.

The article entitled “Accurate Evaluation of Feature Contributions for Sentinel Lymph Node Status Classification in Breast Cancer” by Lombardi et al. has considered the sentinel lymph node status in breast cancer, which is an unbalanced classification issue. The authors adopt machine learning techniques to study the effect of different sampling strategies on both performance and feature stability for the prediction of the sentinel lymph node status in breast cancer.

Huyut et al., in “Detection of Risk Predictors of COVID-19 Mortality with Classifier Machine Learning Models Operated with Routine Laboratory Biomarkers”, have determined 34 routine blood values, adopting a statistical approach to find the most accurate classifier to predict mortality in patients suffering from COVID-19. Moreover, the lethal risk level of these values is determined using threshold approaches.

In the paper “Performance Analysis for COVID-19 Diagnosis Using Custom and State-of-the-Art Deep Learning Models”, Nagi et al. evaluated the performance of well-known deep learning models on a larger COVID-19 chest X-ray image dataset for the implementation of a DSS for the accurate diagnosis of the SARS CoV-2 virus.

Rizzi et al. in the study named “A Decision Support System for Melanoma Diagnosis from Dermoscopic Images” developed a new pipeline method which uses computer vision procedure for the detection of melanoma in dermoscopic images. The strategy is inspired by clinical practice but adopts original features linked to both geometric properties and chromatic characteristics. Moreover, a statistical study is performed for the evaluation of image quality based on the method’s performance.

A computer-aided system for the prediction of the risk for second primary skin cancer survivors is developed by Lee et al. in the paper entitled “Clinical Risk Factor Prediction for Second Primary Skin Cancer: A Hospital-Based Cancer Registry”. The authors identify the clinical characteristics of second primary skin cancer and highlight that age, stage, gender, and involvement of the regional lymph nodes are the most significant impact factors for second primary skin cancer.

Du et al., in the paper “The Role of XAI in Advice-Taking from a Clinical Decision Support System: A Comparative User Study of Feature Contribution-Based and Example-Based Explanations”, analyzed the behavior of clinicians during the adoption of machine learning-based DSS in the cases of explanation by feature and explanation by example. They underline that healthcare practitioners have different preferences for explanations, so during the development phase of DSSs, designers should select the method in accordance with the users. Moreover, it should be advisable to equip DSSs with several options of explanations for physicians with different clinical expertise and cognitive styles.

In the article “An Ensemble Framework to Improve the Accuracy of Prediction Using Clustered Random-Forest and Shrinkage Methods”, Farhadi et al. have highlighted that in the field of prediction, all the requisite steps/actions must be taken for computational time reduction, in addition to developing procedures with high accuracy. To achieve this aim, the authors propose a combination of statistical methods and machine-learning algorithms.

Considering the above, it is clear that this Special Issue highlights several DSS applications to detect a variety of pathologies/disorders by means of different techniques and diagnostic investigations.

The analysis of the selected papers and of several studies indicated in the literature shows that the adoption of efficient and effective DSSs aids clinical assessment, makes the diagnosis reproducible by eliminating inter- and intra-observer variabilities, prevents misdiagnosis, reduces mortality rates, decreases the cost of treatment, and facilitates evidence-based decision-making [14,15,16,17]. Despite the manifest benefits of DSS adoption in clinical practice, there are several challenges that hamper their widespread application [18]. Validation in a clinical environment has to be carried out for DSS to have perfect integration into the clinical workflow. Moreover, dissemination of basic knowledge for their use, training of qualified personnel, usability and friendly designed user interfaces are some of the success strategies to be implemented for enhancing DSS clinician acceptance [19,20,21].

In conclusion, future studies and developments of DDSs for disease detection and diagnosis should try to overcome current limitations and extend the scope of their use in different application contexts. Looking towards new and emerging technologies, DSSs will assume an increasingly relevant role in shaping the feature of healthcare by changing the way healthcare services will be delivered to patients.

At the end of this editorial, I am very grateful to all those who have contributed to the realization of the Special Issue. I thank all the authors of the published papers; without them, the Special Issue would not be possible, and I am grateful to all reviewers for their valuable comments and suggestions. Last but not least, a special thanks to the Editor in Chief, the Section Editor in Chief, the Editorial Board, the Special Issue Editor, the Assistant Editor, and the Editorial Office team for their support, collaboration, and useful tips.