This contribution briefly describes the research being carried out in the Computational Intelligence Laboratory of the array of methodologies and applications aimed at leveraging the capability of AI to empower the diagnosis, monitoring, and treatment of various health conditions. Through multifaceted research that covers neuroimaging analysis, acoustic signal processing, and vital parameter monitoring, our goal is to shed light on the potential of AI in enhancing healthcare services. Workshop Proceedings (CILab) of the Department of Computer Science, Uni- fects around 55 million people worldwide, predominantly aims to address current healthcare challenges by devel- treatments focusing more on symptom management than versity of Bari Aldo Moro, we are contributing to this transformation by applying AI to neuroimaging, acoustic signal analysis, and vital signs monitoring. Our work oping innovative and practical AI solutions.
artificial intelligence, explainability, e-health, neuroimaging, acoustic signals, vital parameters CEUR
ceur-ws.org
Integrating Artificial Intelligence (AI) into healthcare is transforming how we diagnose, monitor, and care for ifeld of AI in healthcare. 2. Neuroimaging folio, where the application of AI-driven algorithms plays a crucial role in enabling the early and precise diagnosis of neurological disorders. nEvelop-O (G. Zaza) hallmark of AD.
Our research has advanced the application of
Convolutional Neural Network (CNN) models for the automated
diagnosis of AD, using the strengths of both MRI and PET
scans [
Specifically, our investigation has yielded several key then used to train a supervised 3D CNN for AD detecinsights, highlighting the potential of multi-modal imag- tion. This innovative strategy demonstrated encouraging ing strategies. By analyzing the classification results from outcomes on the OASIS-3 dataset and lessened the devarious model configurations on the OASIS-3 benchmark pendence on extensively annotated datasets, setting the dataset, we discovered that models utilizing 3D inputs stage for more autonomous and quantitative AD detecconsistently outperformed those using 2D inputs, likely tion in clinical practice. Our future endeavors will aim due to the richer spatial information available in 3D scans. to assess this method across broader and more varied Moreover, regardless of whether in 2D or 3D, MRI scans datasets to afirm its diagnostic validity further. significantly surpassed amyloid PET scans in diagnostic performance, emphasizing MRI’s inherent value in AD 2.2. Brain tumor segmentation detection within our study. However, multi-modal strategies, particularly our “fusion” model (shown in Fig. 1), Brain tumor segmentation from MRI scans is crucial for demonstrated a clear advantage, achieving up to 95% ac- accurate diagnosis, treatment planning, and patient moncuracy. This underlines the complementary nature of itoring. Recent strides in deep learning, particularly with MRI and PET scans in AD diagnosis. CNNs, have significantly advanced the automation and
Our adoption of the Grad-CAM technique further al- precision of tumor segmentation. Nevertheless, these lowed us to pinpoint the brain regions most relevant models’ “black-box” nature raises challenges in explainfor classification, ofering valuable insights into the neu- ability—a vital aspect of clinician trust and decisionroanatomical underpinnings of AD. This supports the making. validity of our models and enhances our understanding Graph Neural Networks (GNNs) have recently gained of AD’s neuropathology. attention as a novel approach to medical image analysis,
In another exploratory study [
We introduced a dual-stage deep learning approach, as a promising tool for achieving high precision in brain
combining unsupervised and supervised techniques. Ini- tumor segmentation and providing a pathway to model
tially, a 3D convolutional autoencoder was employed to understandability.
extract low-dimensional representations from FA images In a recent study [
Our exploration highlighted the efectiveness of GNNs designed to monitor BD states while considering the
in medical imaging. It also laid the foundation for future temporal acquisition of acoustic features. DISSFCM, an
research, suggesting potential synergies between GNNs extension of the Semi-Supervised Fuzzy C-Means
(SSand CNNs, such as integrating GNNs with 3D U-Net FCM) algorithm, analyses data chunks sequentially in
architectures, to refine segmentation results further. In near real-time, maintaining historical data insights
withaddition, collaboration with medical experts to examine out extensive storage. It adapts to new information,
refincritical features identified by GNNExplainer could further ing the classification through an increased cluster count
solidify the role of GNNs in clinical practice, combining representing the patient’s condition states. This method
accuracy and explainability in brain tumor management. has proven efective in predicting episodes of health and
illness with as little as 25% labeled data [
DISSFCM operates on labeled prototypes,
summariz3. Acoustic signals ing data clusters for each segment. It generates
membership matrices, clarifying each data point’s cluster
association and facilitating outcome explanation. Initially, we
applied visual analytics for interpretation [
They ofer clear, insightful linguistic descriptions, making the complex data and sparse psychiatric evaluations comprehensible.
The analysis of vocal characteristics from speech samples is an efective approach to identifying conditions associated with mental diseases, notably bipolar disorder (BD). Our research focused on extracting acoustic features from patients with BD using a specialized mobile application, developed at the Department of Afective Disorders, Institute of Psychiatry and Neurology in Warsaw, Poland, under the project “Smartphone-based diagnostics of phase changes in the course of bipolar disorder”.
BD manifests through fluctuating mood states, including euthymia, depression, mixed states, and mania, traditionally diagnosed through regular consultations using standard psychiatric tools like the Hamilton Depression Rating Scale (HDRS) and the Young Mania Rating Scale (YMRS). These instruments allow healthcare providers to detect symptoms and evaluate the intensity of depressive and manic episodes, facilitating precise diagnoses and the development of customized treatment strategies.
Our research examined several critical dimensions of data related to BD, specifically focusing on the importance of continuous monitoring to track temporal fluctuations. We addressed the challenge of missing labels while also handling the uncertainty in labeling due to the inherent ambiguity and variability in data classification.
Moreover, we worked on generating readily understandable explanations of BD state classifications leveraging the availability of multi-layered information.
We designed a versatile, multi-task neural network to
leverage the detailed symptom information captured
during patient assessments. This network is trained to
generate several outputs, each aligning with the various
levels of labels obtained from intermediate assessment
stages. These intermediate outputs fulfill dual roles: they
enhance the model’s overall predictive accuracy and
provide insights into classifying mid-level labels. Our
architecture, designed to handle data with a hierarchical
class structure, is a crucial component of PLENARY
(ex3.1. Monitoring bipolar disorder states Plaining bLack-box modEls in Natural lAnguage thRough
fuzzY linguistic summaries) [
We explored semi-supervised learning algorithms to anxiety, decreased activity, mood changes, disorganizaharness the geometric data properties and the prede- tion, and sleep disorders, among others. The model’s ifned knowledge of patient states. These algorithms have outcomes and explanations focus on the patient’s state
AI to anticipate important diseases, equipping patients and physicians with critical insights for preemptive healthcare management. Herein, we detail our eforts in remote vital parameter estimation and creating eXplainable AI (XAI) models to support medical diagnosis.
These models use vital signs data to aid medical professionals in the early detection of cardiovascular diseases and stress-related conditions. and these specific symptoms. For instance, we found that individuals from the discomfort of traditional contact“Among records that contribute positively to predicting based monitoring, making it a more convenient and usermania, most of them have spectral-related features at friendly option for daily home use. low level” and “Among records that contribute against Our methodological pipeline (shown in Fig. 2(b)) inipredicting decreased activity, most of them have quality- tiates with the detection of the subject’s face, focusing related features at low level”. specifically on the forehead as the region of interest (ROI)
Through rigorous experimental evaluation, we have for signal extraction. The rPPG signal is then processed demonstrated that augmenting model explanations with using Independent Component Analysis (ICA) and Fast fuzzy linguistic summarization—especially those derived Fourier Transform (FFT) to estimate HR and BR. At the from SHAP analyses—significantly enhances understand- same time, SpO2 measurements are derived by applying of the model’s predictions. This approach efectively ing the Beer-Lambert law. For lip color detection, the combines domain-specific knowledge with technical in- system identifies the lip ROI and determines the domsight, providing a comprehensive and accessible explana- inant color using clustering methods. Our contactless tion framework. approach has not only demonstrated measurement accuracy within acceptable ranges for both stationary and minimally moving subjects, but it has also shown supe4. Vital parameters rior performance compared to traditional contact devices, instilling confidence in its reliability and accuracy.
Further enhancements included the addition of new
ROIs and a face-tracking feature to accommodate head
movements, improving usability on mobile devices [
4.1. Contact-less monitoring of vital While traditional machine learning algorithms have sigparameters nificantly aided physicians in diagnosing symptoms early to prevent disease progression, their often opaque nature Our endeavors in vital parameter monitoring have been presents a challenge. These “black box” models deliver concentrated on heart rate, breathing rate, blood oxy- accurate predictions but lack an intuitive explanation gen saturation (SpO2), and systolic and diastolic blood for their results. This makes them less practical in fields pressure—key indicators for cardiovascular health. Tra- where end-users are non-technical professionals, notably ditional methods, like ECG, require direct skin contact, in healthcare. often necessitating cumbersome wearable devices. To Our work concentrates on advancing XAI models, overcome the limitations and discomfort of contact-based which are mainly aimed at supporting medical decisions monitoring, advancements have been made toward de- in cardiovascular disease (CVD) assessment. CVDs are a veloping photoplethysmography (PPG) techniques that primary global health concern, responsible for approxoperate using camera-based systems. However, these can imately 17.9 million deaths annually,1 spanning condibe expensive and not user-friendly for daily home use. tions such as coronary heart disease and stroke. Given
Addressing these challenges, our lab has developed an the multifactorial causes of CVDs, including lifestyle and
innovative, cost-efective approach for monitoring car- genetic predispositions, early intervention and
continudiovascular parameters that seamlessly integrates into ous monitoring of vital signs are crucial to prevention.
everyday living environments [
To bridge the gap between data-driven precision and
expert intuition, we explored neuro-fuzzy systems, which
automate the generation of fuzzy rule-based models
from data, streamlining the otherwise manual and
laborintensive process of rule formation. Our research
demonstrates that models created through neuro-fuzzy
systems maintain accuracy and significantly enhance
interpretability, outperforming manually designed models
in cardiovascular risk prediction [
Expanding beyond cardiovascular health, we have
applied neuro-fuzzy systems to diagnose hypertension and
stress, focusing on minimizing complexity for clearer
understanding. We have balanced accuracy and
interpretability by employing feature selection to refine the
number of relevant indicators and fuzzy rules, making
these models highly practical for real-world medical
applications [
Future AI Research (PE00000013) project, Spoke 6 - Symbiotic AI (CUP H97G22000210007), under the NRRP MUR program funded by NextGenerationEU. Ga.C. acknowledges funding from the European Union PON project Ricerca e Innovazione 2014-2020, D.M. 1062/2021. All authors are members of the INdAM GNCS research group. Ga.C, G.C, and G.V. are members of the CITEL - Centro Interdipartimentale della ricerca in Telemedicina, University of Bari Aldo Moro.