Biology subjects

Grossman, R.

Publications and source records attributed to Grossman, R..

3 recordsLinked to original sources

bioRxiv2021

Vocal markers of Autism Spectrum Disorder: assessing the generalizability of machine learning models

Machine learning (ML) approaches show increasing promise in their ability to identify vocal markers of autism. Nonetheless, it is unclear to what extent such markers generalize to new speech samples collected e.g., using a different speech task or in a different language. In this paper, we systematically assess the generalizability of ML findings across a variety of contexts. We train promising published ML models of vocal markers of autism on novel cross-linguistic datasets following a rigorous pipeline to minimize overfitting, including cross-validated training and ensemble models. We test the generalizability of the models by testing them on i) different participants from the same study, performing the same task; ii) the same participants, performing a different (but similar) task; iii) a different study with participants speaking a different language, performing the same type of task. While model performance is similar to previously published findings when trained and tested on data from the same study (out-of-sample performance), there is considerable variance between studies. Crucially, the models do not generalize well to different, though similar, tasks and not at all to new languages. The ML pipeline is openly shared. Generalizability of ML models of vocal markers of autism is an issue. We outline three recommendations for strategies researchers could take to be more explicit about generalizability and improve it in future studies. LAY SUMMARYMachine learning approaches promise to be able to identify autism from voice only. These models underestimate how diverse the contexts in which we speak are, how diverse the languages used are and how diverse autistic voices are. Machine learning approaches need to be more careful in defining their limits and generalizability.

bioinformatics↗

bioRxiv2021

Towards a cumulative science of prosody in autism: a cross-linguistic meta-analysis-based investigation of acoustic markers in American and Danish autistic children

Acoustic atypicalities in speech production are argued to be potential markers of clinical features in Autism Spectrum Disorder (ASD). A recent meta-analysis highlighted shortcomings in the field, in particular small sample sizes and study heterogeneity (Fusaroli et al., 2017). We showcase a cumulative (i.e., explicitly building on previous studies both conceptually and statistically) yet self-correcting (i.e., critically assessing the impact of cumulative statistical techniques) approach to prosody in ASD to overcome these issues. We relied on the recommendations contained in the meta-analysis to build and analyze a cross-linguistic corpus of multiple speech productions in 77 autistic and 72 neurotypical children and adolescents (>1000 recordings in Danish and US English). We used meta-analytically informed and skeptical priors, with informed priors leading to more generalizable inference. We replicated findings of a minimal cross-linguistically reliable distinctive acoustic profile for ASD (higher pitch and longer pauses) with moderate effect sizes. We identified novel reliable differences between the two groups for normalized amplitude quotient, maxima dispersion quotient, and creakiness. However, the differences were small, and there is likely no one acoustic profile characterizing all autistic individuals. We identified reliable relations of acoustic features with individual differences (age, gender), and clinical features (speech rate and ADOS sub-scores). Besides cumulatively building our understanding of acoustic atypicalities in ASD, the study shows how to use systematic reviews and meta-analyses to guide the design and analysis of follow-up studies. We indicate future directions: larger and more diverse cross-linguistic datasets, focus on heterogeneity, self-critical cumulative approaches and open science. Lay SummaryAutistic individuals are reported to speak in distinctive ways. Distinctive vocal production can affect social interactions and social development and could represent a noninvasive way to support the assessment of ASD. We systematically checked whether acoustic atypicalities highlighted in previous articles could be actually found across multiple recordings and two languages. We find a minimal acoustic profile of ASD: higher pitch, longer pauses, increased hoarseness and creakiness of the voice. However, there is much individual variability (by age, sex, language, and clinical characteristics). This suggests that the search for one common "autistic voice" might be naive and more fine-grained approaches are needed.

bioinformatics↗

bioRxiv2020

The Impact of Digital Histopathology Batch Effect on Deep Learning Model Accuracy and Bias

The Cancer Genome Atlas (TCGA) is one of the largest biorepositories of digital histology. Deep learning (DL) models have been trained on TCGA to predict numerous features directly from histology, including survival, gene expression patterns, and driver mutations. However, we demonstrate that these features vary substantially across tissue submitting sites in TCGA for over 3,000 patients with six cancer subtypes. Additionally, we show that histologic image differences between submitting sites can easily be identified with DL. This site detection remains possible despite commonly used color normalization and augmentation methods, and we quantify the digital image characteristics constituting this histologic batch effect. As an example, we show that patient ethnicity within the TCGA breast cancer cohort can be inferred from histology due to site-level batch effect, which must be accounted for to ensure equitable application of DL. Batch effect also leads to overoptimistic estimates of model performance, and we propose a quadratic programming method to guide validation that abrogates this bias.

bioinformatics↗