Facilitating clinically relevant skin tumor diagnostics with spectroscopy-driven machine learning.

In the dawning era of artificial intelligence (AI), health care stands to undergo a significant transformation with the increasing digitalization of patient data. Digital imaging, in particular, will serve as an important platform for AI to aid decision making and diagnostics. A growing number of studies demonstrate the potential of automatic pre-surgical skin tumor delineation, which could have tremendous impact on clinical practice. However, current methods rely on having ground truth images in which tumor borders are already identified, which is not clinically possible. We report a novel approach where hyperspectral images provide spectra from small regions representing healthy tissue and tumor, which are used to generate prediction maps using artificial neural networks (ANNs), after which a segmentation algorithm automatically identifies the tumor borders. This circumvents the need for ground truth images, since an ANN model is trained with data from each individual patient, representing a more clinically relevant approach.

The first aphasia screening test in Hungarian: A preliminary study on validity and diagnostic accuracy.

The Hungarian Aphasia Screening Test (HAST) is a newly developed diagnostic tool for detecting post-stroke aphasia in clinical settings, and for differentiating between stroke patients with and without aphasia. The HAST was developed by our team and has not been published in English yet. In Hungarian, to date, there is no aphasia screening test with reported psychometric properties available. This study aims to present the main characteristics of the HAST and to evaluate its validity, internal consistency, and diagnostic accuracy. The HAST comprises five subtests (maximum score: 20) and takes 5-10 minutes to administer. We administered the HAST to 40 stroke patients with aphasia, 26 stroke patients without aphasia, and 51 healthy control participants to evaluate the test's construct validity, convergent validity, and internal consistency, as well as its sensitivity and specificity. We used the Western Aphasia Battery (WAB) as a reference test. With a cut-off score of 17, the HAST showed high diagnostic accuracy (sensitivity: 92.5%, specificity: 88.5%). Its construct validity was good; we identified one component in the test, and moderate-to-strong positive correlations across most of its subtests (mean Spearman r = 0.56). Convergent validity of the HAST was satisfying, reflected by the moderate-to-strong positive correlations between subtests of the HAST and subtests of the WAB (Spearman r = 0.50-0.86). The correlation between the HAST total score and the WAB aphasia quotient was high (Spearman r = 0.86). Despite the small number of items within tasks, all subtests showed acceptable internal consistency (mean Cronbach's α = 0.74). Our preliminary results suggest that the HAST is a valid, accurate, and clinically feasible test to detect post-stroke aphasia and to identify patients who require a more detailed assessment of their language skills. In addition, it reliably identifies not only the presence but also the severity of aphasia, thus, it might be a good candidate for monitoring patient progress.

[The effects of demographic and clinical factors on the severity of poststroke aphasia]. / A demográfiai és klinikai faktorok hatása a poststroke afázia súlyosságára.

Background and purpose:

Poststroke aphasia severity is related to several demographic, lesion-specific, and clinical factors. However, results about the importance of these factors are controversial. The aim of the current study was to investigate the effects of demographic and clinical factors on aphasia severity as well as on expressive and receptive language skills in a sample of Hungarian-speaking people with aphasia. 

Ninety-four people with aphasia with mainly unilateral left-hemisphere stroke (87.88%) participated. We used multiple stepwise linear regression to investigate the relationships between potential predictors – i.e., sex, education, time postonset, etiology, lesion localisation, pathological changes in the brain caused by small vessel disease, and other neurogenic communication disorders/swallowing disorders – and language outcome. As outcome variables, we used the total score, the receptive score, and the expressive score of the Hungarian Aphasia Screening Test.

Less years of education, pathological changes in the brain (white matter hypodensities/hyperintensities and subcortical lacunar lesions) and apraxia of speech were associated with more severe aphasia. Pathological changes and apraxia of speech were also associated with lower expressive language skills in aphasia. Finally, education, pathological changes, and sex predicted receptive language skills in poststroke aphasia (p < 0.05 in all three models).  

Pathological changes, apraxia of speech, education, and sex may affect language outcome in poststroke aphasia. We discuss our findings in light of the results of previous studies. 

Digital Papillary Adenocarcinoma in Nonacral Skin: Clinicopathologic and Genetic Characterization of 5 Cases.

Digital papillary adenocarcinoma (DPA) is a rare sweat gland neoplasm that has exceptionally been reported outside acral locations. Recently, human papillomavirus 42 was identified as the main oncogenic driver of DPA. Herein, we report 5 tumors arising in extra-acral locations predominantly in the female anogenital skin. Four patients were female and 1 patient was male. The mean age at the diagnosis time was 65 years (range: 55 to 82 y). Tumors were located on the vulva (n=3), perianal area (n=1), and forearm (n=1). Histologically, all tumors were lobular and mainly solid and composed of sheets of cells with rare focal papillae and frequent glandular structures in a "back-to-back" pattern and lined by atypical basophilic cells. Immunohistochemistry showed diffuse positivity for SOX10. Epithelial membrane antigen and carcinoembryonic antigen highlighted the luminal cells and staining for p63 and p40 revealed a consistent and continuous myoepithelial component around glandular structures. Follow-up was available in 3 cases (mean duration: 12 mo [range: 8 to 16 mo]). One patient developed local recurrence and 1 experienced regional lymph node metastases. HPV Capture Next-generation sequencing revealed the presence of the HPV42 genome in all samples. Viral reads distributions were compatible in the 5 cases with an episomal nature of the viral genome, with a recurrent deletion in the E1 and/or E2 open reading frames. In conclusion, this study demonstrates that digital DPA may rarely present in nonacral locations mainly in the female anogenital area, usually with a more solid pattern as compared with those cases presenting on the digits and it is also associated with HPV42.

Reading fluency and statistical learning across modalities and domains: Online and offline measures.

The vulnerability of statistical learning has been demonstrated in reading difficulties in both the visual and acoustic modalities. We examined segmentation abilities of Hungarian speaking adolescents with different levels of reading fluency in the acoustic verbal and visual nonverbal domains. We applied online target detection tasks, where the extent of learning is reflected in differences between reaction times to predictable versus unpredictable targets. Explicit judgments of well-formedness were also elicited in an offline two-alternative forced choice (2AFC) task. Learning was evident in both the acoustic verbal and visual nonverbal tasks, both in online and offline measures, but learning effects were larger both in online and offline tasks in the verbal acoustic condition. We haven't found evidence for a significant relationship between statistical learning and reading fluency in adolescents in either modality. Together with earlier findings, these results suggest that the relationship between reading and statistical learning is dependent on the domain, modality and nature of the statistical learning task, on the reading task, on the age of participants, and on the specific language. The online target detection task is a promising tool which can be adapted to a wider set of tasks to further explore the contribution of statistical learning to reading acquisition in participants from different populations.

Modality, presentation, domain and training effects in statistical learning.

While several studies suggest that the nature and properties of the input have significant effects on statistical learning, they have rarely been investigated systematically. In order to understand how input characteristics and their interactions impact statistical learning, we explored the effects of modality (auditory vs. visual), presentation type (serial vs. simultaneous), domain (linguistic vs. non-linguistic), and training type (random, starting small, starting big) on artificial grammar learning in young adults (N = 360). With serial presentation of stimuli, learning was more effective in the auditory than in the visual modality. However, with simultaneous presentation of visual and serial presentation of auditory stimuli, the modality effect was not present. We found a significant domain effect as well: a linguistic advantage over nonlinguistic material, which was driven by the domain effect in the auditory modality. Overall, the auditory linguistic condition had an advantage over other modality-domain types. Training types did not have any overall effect on learning; starting big enhanced performance only in the case of serial visual presentation. These results show that input characteristics such as modality, presentation type, domain and training type influence statistical learning, and suggest that their effects are also dependent on the specific stimuli and structure to be learned.

[Hungarian Aphasia Screening Test: a new rapid screening tool for detecting aphasia in stroke patients]. / Magyar Afázia Szuroteszt: új szuroeljárás a poststroke aphasia gyors felismerésére.

INTRODUCTION: Although aphasia screening tools are widely used internationally, in Hungarian, to date, there have been no aphasia screening tests with reported psychometric properties available. OBJECTIVE: (1) To provide an overview of the international practice and the tools of post-stroke aphasia screening. (2) To present a recently developed aphasia screening test in Hungarian. METHOD: The Hungarian Aphasia Screening Test comprises five tasks assessing speech comprehension and production, and takes 5-10 minutes to complete. The test is easy to score and needs little equipment, therefore it can be useful in any clinical setting that allows for quick test administration and diagnostics. Based on results of two studies, we evaluate the psychometric properties and the diagnostic accuracy of the test, and determine three impairment severity ranks (mild, moderate, severe aphasia) for the test. The severity ranks were defined based on results of 70 people with post-stroke aphasia (35 women; average age: 61.4 years; average postonset: 12.6 months). RESULTS: The test has high diagnostic accuracy, a sensitivity of 92.5% and a specificity of 88.5%. It shows acceptable structural and concurrent (MR = 0.68) validity as well as good internal consistency (Mα = 0.74). The cut-off score for diagnosing aphasia is 17. DISCUSSION: The test can differentiate stroke patients with and without aphasia. With the use of the test, patients' aphasia can be rated as mild (score 14-17), moderate (score 9-13), or severe (score 0-8). Our results suggest that it is a reliable and valid screening tool. CONCLUSION: The test can detect the presence and quantify the severity of aphasia. We discuss the importance of aphasia screening in clinical settings and of the use of the Hungarian Aphasia Screening Test in the context of the international practices. Orv Hetil. 2022; 163(50): 2000-2008.

The role of bile acid measurement in the management of intrahepatic cholestasis of pregnancy / Az epesavmérés szerepe a súlyos terhességi cholestasis szülészeti ellátásában

Introduction: Intrahepatic cholestasis of pregnancy complicates 1% of pregnancies. It increases the risk of severe fetal complications significantly, including preterm delivery and stillbirth. Objective: To summarize our experience with serum total bile acid level measurement that has recently become available for clinical routine in Hungary, and to present the way of gestational cholestasis care at our university. Patients and method: In a retrospective case series, we analyse the data of 12 patients suffering from severe cholestasis of pregnancy treated between September 2020 and September 2021 at the Department of Obstetrics and Gynecology, University of Debrecen. We also determine the statistical correlation between bile acid, transaminase and bilirubin levels in severe cholestasis. Results: 1258 serum samples of 758 patients were measured. 5 of them (0.7% of all cases, 6.4% of cholestasis cases) had severe (total bile acid 40-99 mu mol/L), 7 (0.9% of all cases and 9.0% of cholestasis cases) had very severe (total bile acid >= 100 mu mol/L) disease. The average age of the 12 cases was 30.6 (21-43) years, 7 of them were primigravid. 5 of the patients had a predisposing disease in their history. 6/12 patients received ursodeoxycholic acid treatment, resulting in significant decrease in the bile acid concentrations. Bile acid and GOT (R-2 = 0,14) and bile acid and GPT (R-2 = 0,17) correlations were found to be week in severe cholestasis (n = 45). Postpartum bile acid levels showed rapid improvement. So far, 11 of the patients have delivered and 13 neonates were born, 2/12 were multiple pregnancies. Average gestational age at delivery was 37 (33-40) weeks. 3/11 (27%) were preterm deliveries. 7/8 (88%) of term deliveries were induced. Elective cesarean delivery was not indicated in any of the cases, and in only 2/11 (18%) of the cases did emergency cesarean sections become necessary during labour. No stillbirth occurred. Conclusion: Serum total bile acid measurement is an effective tool in the diagnosis and follow-up of intrahepatic cholestasis of pregnancy, and is inevitable for the protocoll-based obstetrical management of patients. We also present the local protocol of our Department for the management of obstetrical cholestasis.

Photoacoustic imaging of periorbital skin cancer ex vivo: unique spectral signatures of malignant melanoma, basal, and squamous cell carcinoma.

Radical excision of periorbital skin tumors is difficult without sacrificing excessive healthy tissue. Photoacoustic (PA) imaging is an emerging non-invasive biomedical imagi--ng modality that has potential for intraoperative micrographic control of surgical margins. This is the first study to assess the feasibility of PA imaging for the detection of periocular skin cancer. Eleven patients underwent surgical excision of periocular skin cancer, one of which was a malignant melanoma (MM), eight were basal cell carcinomas (BCCs), and two squamous cell carcinomas (SCCs). Six tumors were located in the eyelid, and five in periocular skin. The excised samples, as well as healthy eyelid samples, were scanned with PA imaging postoperatively, using 59 wavelengths in the range 680-970 nm, to generate 3D multispectral images. Spectral unmixing was performed using endmember spectra for oxygenated and deoxygenated Hb, melanin, and collagen, to iden--tify the chromophore composition of tumors and healthy eyelid tissue. After PA scanning, the tumor samples were examined histopathologically using standard hematoxylin and eosin staining. The PA spectra of healthy eyelid tissue were dominated by melanin in the skin, oxygenated and deoxygenated hemoglobin in the orbicularis oculi muscle, and collagen in the tarsal plate. Multiwavelength 3D scanning provided spectral information on the three tumor types. The spectrum from the MM was primarily reconstructed by the endmember melanin, while the SCCs showed contributions primarily from melanin, but also HbR and collagen. BCCs showed contributions from all four endmembers with a predominance of HbO2 and HbR. PA imaging may be used to distinguish different kinds of periocular skin tumors, paving the way for future intraoperative micrographic control.

Merkel Cell Polyomavirus‒Negative Merkel Cell Carcinoma Originating from In Situ Squamous Cell Carcinoma: A Keratinocytic Tumor with Neuroendocrine Differentiation.

Although virus-negative Merkel cell carcinoma (MCC) is characterized by a high frequency of UV-induced mutations, the expression of two viral oncoproteins is regarded as a key mechanism driving Merkel cell polyomavirus‒positive MCC. The cells in which these molecular events initiate MCC oncogenesis have yet not been identified for both MCC subsets. A considerable proportion of virus-negative MCC is found in association with squamous cell carcinoma (SCC), suggesting (i) coincidental collision, (ii) one providing a niche for the other, or (iii) one evolving from the other. Whole-exome sequencing of four combined tumors consisting of SCC in situ and Merkel cell polyomavirus‒negative MCC showed many mutations shared between SCC and MCC in all cases, indicating a common ancestry and thereby a keratinocytic origin of these MCCs. Moreover, analyses of the combined cases as well as of pure SCC and MCC suggest that RB1 inactivation in SCC facilitates MCC development and that epigenetic changes may contribute to the SCC/MCC transition.

Online Statistical Learning in Developmental Language Disorder.

Purpose: The vulnerability of statistical learning (SL) in developmental language disorder (DLD) has mainly been demonstrated with metacognitive offline measures which give little insight into the more specific nature and timing of learning. Our aims in this study were to test SL in children with and without DLD with both online and offline measures and to compare the efficiency of SL in the visual and acoustic modalities in DLD. Method: We explored SL in school-age children with and without DLD matched on age and sex (n = 36). SL was investigated with the use of acoustic verbal and visual nonverbal segmentation tasks relying on online (reaction times and accuracy) and offline (two-alternative forced choice, 2AFC and production) measures. Results: In online measures, learning was evident in both groups in both the visual and acoustic modalities, while offline measures showed difficulties in DLD. The visual production task showed a significant learning effect in both groups, while the visual two-alternative forced choice (2AFC) and the two acoustic offline tasks only showed evidence of learning in the control group. The comparison of learning indices revealed an SL impairment in DLD, which is present in both modalities. Conclusions: Our findings suggest that children with DLD are comparable to typically developing (TD) children in their ability to extract acoustic verbal and visual nonverbal patterns that are cued only by transitional probabilities in online tasks, but they show impairments on metacognitive measures of learning. The pattern of online and offline measures implies that online tests can be more sensitive and valid indices of SL than offline tasks, and the combined use of different measures provides a better picture of learning efficiency, especially in groups where metacognitive tasks are challenging.

Automatic threshold selection algorithm to distinguish a tissue chromophore from the background in photoacoustic imaging.

The adaptive matched filter (AMF) is a method widely used in spectral unmixing to classify different tissue chromophores in photoacoustic images. However, a threshold needs to be applied to the AMF detection image to distinguish the desired tissue chromophores from the background. In this study, we propose an automatic threshold selection (ATS) algorithm capable of differentiating a target from the background, based on the features of the AMF detection image. The mean difference between the estimated thickness, using the ATS algorithm, and the known values was 0.17 SD (0.24) mm for the phantom inclusions and -0.05 SD (0.21) mm for the tissue samples of malignant melanoma. The evaluation shows that the thickness and the width of the phantom inclusions and the tumors can be estimated using AMF in an automatic way after applying the ATS algorithm.

Comparison of photoacoustic imaging and histopathological examination in determining the dimensions of 52 human melanomas and nevi ex vivo.

Surgical excision followed by histopathological examination is the gold standard for the diagnosis and staging of melanoma. Reoperations and unnecessary removal of healthy tissue could be reduced if non-invasive imaging techniques were available for presurgical tumor delineation. However, no technique has gained widespread clinical use to date due to shallow imaging depth or the absence of functional imaging capability. Photoacoustic (PA) imaging is a novel technology that combines the strengths of optical and ultrasound imaging to reveal the molecular composition of tissue at high resolution. Encouraging results have been obtained from previous animal and human studies on melanoma, but there is still a lack of clinical data. This is the largest study of its kind to date, including 52 melanomas and nevi. 3D multiwavelength PA scanning was performed ex vivo, using 59 excitation wavelengths from 680 nm to 970 nm. Spectral unmixing over this broad wavelength range, accounting for the absorption of several tissue chromophores, provided excellent contrast between healthy tissue and tumor. Combining the results of spectral analysis with spatially resolved information provided a map of the tumor borders in greater detail than previously reported. The tumor dimensions determined with PA imaging were strongly correlated with those determined by histopathological examination for both melanomas and nevi.

Does rhythmic priming improve grammatical processing in Hungarian-speaking children with and without developmental language disorder?

Research has described several features shared between musical rhythm and speech or language, and experimental studies consistently show associations between performance on tasks in the two domains as well as impaired rhythm processing in children with language disorders. Motivated by these results, in the current study our first aim was to explore whether a short exposure to a regular musical rhythm (i.e., rhythmic priming) can improve subsequent grammatical processing in preschool-aged Hungarian-speaking children with and without developmental language disorder (DLD). Second, we investigated whether rhythmic priming is specific to grammar processing by assessing priming in two additional domains: a linguistic but non-grammatical task (picture naming) and a non-linguistic task (nonverbal Stroop task). Third, to confirm that the rhythmic priming effect originates from the facilitating effect of the regular rhythm and not the negative effect of the control condition, we added a third condition, silence, for all the three tasks. Both groups of children showed better performance on the grammaticality judgment task in the regular compared to both the irregular and the silent conditions but no such effect appeared in the non-grammatical and non-linguistic tasks. These results suggest that (1) rhythmic priming can improve grammatical processing in Hungarian, a language with complex morphosyntax, both in children with and without DLD, (2) the effect is specific to grammar and (3) is a result of the facilitating effect of the regular rhythm. These results could motivate further research about integrating rhythmic priming into traditional speech-language therapy. A video abstract of this article can be viewed at https://youtu.be/zKzGuIjZyvU.

Statistical Learning and the Effect of Starting Small in Developmental Dyslexia.

Purpose Impairments in statistical learning abilities of individuals with developmental dyslexia (DD) have been demonstrated in word segmentation and in visual artificial grammar learning (AGL) tasks, but so far, little attention has been devoted to the AGL abilities of this population in the acoustic verbal domain. This study aimed to test whether adolescents with dyslexia have difficulties in extracting abstract patterns from auditory sequences of nonsense syllables based on a finite state grammar relative to typically developing (TD) peers. We also tested whether incremental presentation of stimuli of different lengths (starting small) has a facilitating effect on learning complex structures in dyslexia (and in TD) as opposed to presenting strings in random order. Method Thirty-one adolescents with DD and 31 age-matched control participants completed an AGL task. Participants passively listened to acoustic sequences of nonsense syllables generated by an artificial grammar in the training phase. In the test phase, they were presented with pairs of novel grammatical and nongrammatical sequences and were required to decide which member of a sequence pair was more similar to the material heard during training. Results Performance levels and the proportion of learners were smaller in participants with DD than in the control group. While the starting small effect was nominally present both in performance levels and in the number of learners in participants with DD, but not in the group with TD, the presentation of strings in incremental order did not statistically improve learning performance in either group. Conclusion Our results suggest that (a) statistical learning of abstract sequences in the acoustic domain is less efficient in people with dyslexia than in TD controls and (b) while incremental presentation of stimuli of different length did not improve learning in our study, the observed pattern of results suggests that the effects of different training designs should be explored further in developmental disorders.

Tracking statistical learning online: Word segmentation in a target detection task.

Despite the essential role of statistical learning in shaping human behavior, there are still controversies concerning its measurement. In this paper, we present a novel online target-detection task in an acoustic word segmentation paradigm, which is able to track the process of learning and does not build on deliberation and decision making. Beside testing the novel online task, we also examined its relationship with two offline measures: the traditional two-alternative forced choice (2AFC) task, and the statistically-induced chunking recall (SICR) task (Isbilen et al., 2017). Participants showed a significant learning effect on the online task, reflected in the decrease of reaction times during training and in the differences between reaction times to predictable versus unpredictable targets. Online learning scores correlated with the 2AFC scores, but this association was only present when participants did not have explicit knowledge about stimuli. SICR scores were not associated with any of the other measures. The internal consistency was higher for online learning measures than for the other two tasks. These findings show that the online target detection task is a good tool for assessing statistical learning, and invite further research on its psychometric properties.

[The Comprehensive Aphasia Test in Hungarian]. / CAT-H ­ új eljárás az afázia magyar nyelvû diagnosztikájában.

BACKGROUND AND PURPOSE: In this paper we present the Comprehensive Aphasia Test-Hungarian (CAT-H; Zakariás and Lukács, in preparation), an assessment tool newly adapted to Hungarian, currently under standardisation. The test is suitable for the assessment of an acquired language disorder, post-stroke aphasia. The aims of this paper are to present 1) the main characteristics of the test, its areas of application, and the process of the Hungarian adaptation and standardisation, 2) the first results from a sample of Hungarian people with aphasia and healthy controls. METHODS: Ninety-nine people with aphasia, mostly with unilateral, left hemisphere stroke, and 19 neurologically intact control participants were administered the CAT-H. In addition, we developed a questionnaire assessing demographic and clinical information. The CAT-H consists of two parts, a Cognitive Screening Test and a Language Test. RESULTS: People with aphasia performed significantly worse than the control group in all language and almost all cognitive subtests of the CAT-H. Consistent with our expectations, the control group performed close to ceiling in all subtests, whereas people with aphasia exhibited great individual variability both in the language and the cognitive subtests. In addition, we found that age, time post-onset, and type of stroke were associated with cognitive and linguistic abilities measured by the CAT-H. CONCLUSION: Our results and our experiences clearly show that the CAT-H provides a comprehensive profile of a person's impaired and intact language abilities and can be used to monitor language recovery as well as to screen for basic cognitive deficits in aphasia. We hope that the CAT-H will be a unique resource for rehabilitation professionals and aphasia researchers in aphasia assessment and diagnostics in Hungary.

Sequence in a sequence: Learning of auditory but not visual patterns within a multimodal sequence.

The current study investigates whether a unimodal visual and a unimodal auditory sequence is learned separately in a multimodal learning situation. In two experiments participants faced a modified version of the Serial Reaction-Time task, in which auditory and visual elements followed each other, forming a multimodal sequence as well as two unimodal sequences. Learning of both the multimodal and the unimodal sequences were tested. Results showed evidence of multimodal sequence learning. The unimodal sequence formed by the auditory stimuli was also learned, while participants did not acquire the concurrent visual sequence. The experiments argue for a domain- and modality-general sequence learning mechanism that is more sensitive to auditory and response-based information than to visual information.

Word Retrieval Difficulties and Cognitive Control in Specific Language Impairment.

Purpose The study aims to test whether children with specific language impairment (SLI) show weaknesses in word retrieval and cognitive control and to find out whether impairments in the 2 domains are associated. Method Thirty-one children with SLI (age: M = 8;11 years;months, SD = 1;1) and 31 age- and IQ-matched typically developing children completed 2 word retrieval tasks and 3 cognitive control tasks. Word retrieval was assessed with a fluency task and a rapid automatized naming (RAN) task, whereas cognitive control was measured with a backward digit span, an n-back task, and a Stroop task. Results We found weaker performance in the SLI group than in the typically developing group in all the fluency conditions and in the size-color-shape RAN as well as on the backward digit span and n-back tasks. Performance on the letter fluency task was associated with backward digit span, whereas size-color-shape RAN performance was associated with backward digit span and n-back scores. Testing the relative contribution of the 3 cognitive control tasks together with verbal short-term memory span and vocabulary size showed that performance on both word retrieval tasks was best explained by nonword repetition and backward digit span measures. Conclusions These results suggest that both word retrieval and cognitive control are impaired in SLI and weaker cognitive control may contribute to word retrieval problems, although weaker short-term memory also has a crucial contribution to word retrieval difficulties during these tasks. If further research confirms this association with cognitive control, training of this ability should be included in the therapy of at least some children with SLI.