Deep learning and machine learning-based voice analysis for the detection of COVID-19: A proposal and comparison of architectures

Costantini, Giovanni; Cesarini, Valerio; Robotti, Carlo; Benazzo, Marco; Pietrantonio, Filomena; Girolamo, Stefano Di; Pisani, Antonio; Canzi, Pietro; Mauramati, Simone; Bertino, Giulia; Cassaniti, Irene; Baldanti, Fausto; Saggio, Giovanni

doi:10.1016/j.knosys.2022.109539

Cited by 28 publications

(12 citation statements)

References 56 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…ML algorithms also are proven to be more reliable, offering more consistent results. We would like to stress that most of our previous work within the same context point to the same conclusion, even when using transfer learning and comparing architectures [ 20 , 24 ]. Moreover, the plethora of studies involving voice analysis for PD, albeit showing a trend towards the usage of CNNs in the last few years, still achieve equally relevant results with ML methods [ 52 ].…”

Section: Discussionmentioning

confidence: 65%

“…Overall, these results show how traditional ML methodologies still hold a relevant place for highly complex tasks such as voice analysis with low-cardinality datasets; on a side note, as limited as the study population might be, this remains a work involving one the biggest datasets for PD detection to-date [ 91 ]. Thus, as many studies and results such as [ 20 , 92 ] and [ 93 ] suggest, ML algorithms can still provide significant results, sometimes improving the state-of-the-art diagnosis, if carefully fine-tuned and applied to the correct features.…”

Section: Discussionmentioning

confidence: 99%

“…The CNN architecture was built with the aims of avoiding overfitting and maximizing performance with a good compromise in net size. Significantly, in previous works with similar tasks (voice analysis for pathology detection, especially directed towards COVID-19), we experimented with transfer learning, using “common” nets such as AlexNet or ResNet [ 89 ], but observed little to no improvement in accuracy with respect to lighter, custom-made nets [ 20 ]. We thus preferred in the current study to implement a model trained from scratch which could be the basis for future implementations of custom CNN models for the analysis of vocal tasks.…”

Section: Methodsmentioning

confidence: 99%

“…Properly validated AI tools can reduce the possible subjectivity bias and “enrich” the scale when applied to the vocal test (even enabling daily evaluations), as different studies have already demonstrated [ 11 , 12 , 20 , 21 , 22 ]. However, the human voice can be potentially influenced by other issues ranging from environmental conditions to subject-specific characteristics [ 23 , 24 , 25 ], so that other forms of evidence are mandatory.…”

Section: Introductionmentioning

confidence: 99%

“…Other than being a reliable means to non-empirically quantify voice impairment in diseases that affect phonatory production, voice analysis is also a completely non-invasive, low-cost and pseudo-real-time solution for deploying telemedicine assessments. Voice-based AI solutions have been successfully experimentally investigated and employed in other medical fields such as dysphonia [ 31 , 32 , 33 ], COVID-19 and pulmonary diseases [ 20 , 22 , 34 , 35 ], and even emotion and stress recognition [ 24 , 36 ].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Artificial Intelligence-Based Voice Assessment of Patients with Parkinson’s Disease Off and On Treatment: Machine vs. Deep-Learning Comparison

Costantini

Cesarini

Leo

et al. 2023

Sensors

Self Cite

View full text Add to dashboard Cite

Parkinson’s Disease (PD) is one of the most common non-curable neurodegenerative diseases. Diagnosis is achieved clinically on the basis of different symptoms with considerable delays from the onset of neurodegenerative processes in the central nervous system. In this study, we investigated early and full-blown PD patients based on the analysis of their voice characteristics with the aid of the most commonly employed machine learning (ML) techniques. A custom dataset was made with hi-fi quality recordings of vocal tasks gathered from Italian healthy control subjects and PD patients, divided into early diagnosed, off-medication patients on the one hand, and mid-advanced patients treated with L-Dopa on the other. Following the current state-of-the-art, several ML pipelines were compared usingdifferent feature selection and classification algorithms, and deep learning was also explored with a custom CNN architecture. Results show how feature-based ML and deep learning achieve comparable results in terms of classification, with KNN, SVM and naïve Bayes classifiers performing similarly, with a slight edge for KNN. Much more evident is the predominance of CFS as the best feature selector. The selected features act as relevant vocal biomarkers capable of differentiating healthy subjects, early untreated PD patients and mid-advanced L-Dopa treated patients.

show abstract

Section: Discussionmentioning

confidence: 65%

Section: Discussionmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Artificial Intelligence-Based Voice Assessment of Patients with Parkinson’s Disease Off and On Treatment: Machine vs. Deep-Learning Comparison

Costantini

Cesarini

Leo

et al. 2023

Sensors

Self Cite

View full text Add to dashboard Cite

show abstract

The Human Digi-real Duality

Saggio

2024

SN COMPUT. SCI.

View full text Add to dashboard Cite

Current technologies allow acquiring whatever amount of data (even big data), from whatever system (object, component, mechanism, network, implant, machinery, structure, asset, etc.), during whatever time lapse (secs, hours, weeks, years). Therefore, potentially it is possible to fully characterize any system for any time we need, with the possible consequence of creating a virtual copy, namely the digital twin (DT) of the system. When technology of DT meets an augmented reality scenario, the augmented digital twin (ADT) arises, when DT meets an artificial intelligence environment, the intelligent digital twin (IDT) arises. DTs, ADTs and IDTs are successfully adopted in electronics, mechanics, chemistry, manufacturing, science, sport, and more, but when adopted for the human body it comes out the human digital twin (HDT) or alternatively named virtual human simulator (VHS). When the VHS incorporates information from surroundings (other VHSs and environment), taking a cue from the particle-wave duality (the mix of matter and energy), we can name this super-VHS as the human digi-real duality (HDRD). This work is focused on defining the aforementioned acronyms, on evidencing their differences, advantages and successful case adoptions, but highlighting technology limits too, and on foreseeing new and intriguing possibilities.

show abstract

Embracing the promise of artificial intelligence to improve patient care in movement disorders

Bhidayasiri,

Goetz

2024

Handbook of Digital Technologies in Movement Disorders

View full text Add to dashboard Cite

Deep learning and machine learning-based voice analysis for the detection of COVID-19: A proposal and comparison of architectures

Cited by 28 publications

References 56 publications

Artificial Intelligence-Based Voice Assessment of Patients with Parkinson’s Disease Off and On Treatment: Machine vs. Deep-Learning Comparison

Artificial Intelligence-Based Voice Assessment of Patients with Parkinson’s Disease Off and On Treatment: Machine vs. Deep-Learning Comparison

The Human Digi-real Duality

Embracing the promise of artificial intelligence to improve patient care in movement disorders

Contact Info

Product

Resources

About