In clinical practice, several standardized neuropsychological tests have been designed to assess and monitor the neurocognitive status of patients with neurodegenerative diseases such as Alzheimer’s disease. Important research efforts have been devoted so far to the development of multivariate machine learning models that combine the different test indexes to predict the diagnosis and prognosis of cognitive decline with remarkable results. However, less attention has been devoted to the explainability of these models. In this work, we present a robust framework to (i) perform a threefold classification between healthy control subjects, individuals with cognitive impairment, and subjects with dementia using different cognitive indexes and (ii) analyze the variability of the explainability SHAP values associated with the decisions taken by the predictive models. We demonstrate that the SHAP values can accurately characterize how each index affects a patient’s cognitive status. Furthermore, we show that a longitudinal analysis of SHAP values can provide effective information on Alzheimer’s disease progression.

Recent works have extensively investigated the possibility to predict brain aging from T1-weighted MRI brain scans. The main purposes of these studies are the investigation of subject-specific aging mechanisms and the development of accurate models for age prediction. Deviations between predicted and chronological age are known to occur in several neurodegenerative diseases; as a consequence, reaching higher levels of age prediction accuracy is of paramount importance to develop diagnostic tools. In this work, we propose a novel complex network model for brain based on segmenting T1-weighted MRI scans in rectangular boxes, called patches, and measuring pairwise similarities using Pearson's correlation to define a subject-specific network. We fed a deep neural network with nodal metrics, evaluating both the intensity and the uniformity of connections, to predict subjects' ages. Our model reaches high accuracies which compare favorably with state-of-the-art approaches. We observe that the complex relationships involved in this brain description cannot be accurately modeled with standard machine learning approaches, such as Ridge and Lasso regression, Random Forest, and Support Vector Machines, instead a deep neural network has to be used.

Malignant pleural mesothelioma (MPM) is a rare neoplasm, mainly caused by asbestos exposure, with a high mortality rate. The management of patients with MPM is controversial due to a long latency period between exposure and diagnosis and because of non-specific symptoms generally appearing at advanced stage of the disease. Breath analysis, aimed at the identification of diagnostic Volatile Organic Compounds (VOCs) pattern in exhaled breath, is believed to improve early detection of MPM. Therefore, in this study, breath samples from 14 MPM patients and 20 healthy controls (HC) were collected and analyzed by Thermal Desorption-Gas Chromatography-Mass Spectrometry (TD-GC/MS). Nonparametric test allowed to identify the most weighting variables to discriminate between MPM and HC breath samples and multivariate statistics were applied. Considering that MPM is an aggressive neoplasm leading to a late diagnosis and thus the recruitment of patients is very difficult, a promising data mining approach was developed and validated in order to discriminate between MPM patients and healthy controls, even if no large population data are available. Three different machine learning algorithms were applied to perform the classification task with a leave-one-out cross-validation approach, leading to remarkable results (Area Under Curve AUC = 93%). Ten VOCs, such as ketones, alkanes and methylate derivates, as well as hydrocarbons, were able to discriminate between MPM patients and healthy controls and for each compound which resulted diagnostic for MPM, the metabolic pathway was studied in order to identify the link between VOC and the neoplasm. Moreover, five breath samples from asymptomatic asbestos-exposed persons (AEx) were exploratively analyzed, processed and tested by the validated statistical method as blinded samples in order to evaluate the performance for the early recognition of patients affected by MPM among asbestos-exposed persons. Good agreement was found between the information obtained by gold-standard diagnostic methods such as computed tomography CT and model output.

Contrast-Enhanced Spectral Mammography (CESM) is a novelty instrumentation for diagnosing of breast cancer, but it can still be considered operator dependent. In this paper, we proposed a fully automatic system as a diagnostic support tool for the clinicians. For each Region Of Interest (ROI), a features set was extracted from low-energy and recombined images by using different techniques. A Random Forest classifier was trained on a selected subset of significant features by a sequential feature selection algorithm. The proposed Computer-Automated Diagnosis system is tested on 48 ROIs extracted from 53 patients referred to Istituto Tumori “Giovanni Paolo II” of Bari (Italy) from the breast cancer screening phase between March 2017 and June 2018. The present method resulted highly performing in the prediction of benign/malignant ROIs with median values of sensitivity and specificity of $87.5\%$ and $91.7\%$ , respectively. The performance was high compared to the state-of-the-art, even with a moderate/marked level of parenchymal background. Our classification model outperformed the human reader, by increasing the specificity over $8\%$ . Therefore, our system could represent a valid support tool for radiologists for interpreting CESM images, both reducing the false positive rate and limiting biopsies and surgeries.

• Multi-site MRI data encode confounding information which may mask case-control differences. • The impact of the NeuroHarmonize method is evaluated in the ASD-control classification. • We verified the successful removal of the site effect by the harmonization protocol. • The increment in the classification performance is quantified after data harmonization. • We identified the anatomical features that contributed to the two-class separation.