Intent discovery is a crucial task in natural language processing, and it is increasingly relevant for various of industrial applications.
Identifying novel, unseen intents from user inputs remains one of the biggest challenges in this field.
Herein, we propose Zero-Shot-BERT-Adapters, a two-stage method for multilingual intent discovery relying on a Transformer architecture,
fine-tuned with Adapters. We train the model for Natural Language Inference (NLI) and later perform unknown intent classification in a zero-shot setting for multiple languages. In our evaluation, we first analyze the quality of the model after adaptive fine-tuning on known classes. Secondly, we evaluate its performance in casting intent classification as an NLI task. Lastly, we test the zero-shot performance of the model on unseen classes, showing how Zero-Shot-BERT-Adapters can effectively perform intent discovery by generating semantically similar intents, if not equal, to the ground-truth ones. Our experiments show how Zero-Shot-BERT-Adapters outperforms various baselines in two zero-shot settings: known intent classification and unseen intent discovery. The proposed pipeline holds the potential for broad application in customer care. It enables automated dynamic triage using a lightweight model that can be easily deployed and scaled in various business scenarios, unlike large language models. Zero-Shot-BERT-Adapters represents an innovative multi-language approach for intent discovery, enabling the online generation of novel intents.
A Python package implementing the pipeline and the new datasets we compiled are available at the following link: https://github.com/GT4SD/zero-shot-bert-adapters.

Zero-Shot-BERT-Adapters: a Zero-Shot Pipeline for Unknown Intent Detection

The tasks of out-of-domain (OOD) intent discovery and generalized intent discovery (GID) aim to extend a closed intent classifier to open-world intent sets, which is crucial to task-oriented dialogue (TOD) systems. Previous methods address them by fine-tuning discriminative models. Recently, although some studies has been exploring the application of large language models (LLMs) represented by ChatGPT to various downstream tasks, it is still unclear for the ability of ChatGPT to discover and incrementally extent OOD intents. In this paper, we comprehensively evaluate ChatGPT on OOD intent discovery and GID, and then outline the strengths and weaknesses of ChatGPT. Overall, ChatGPT exhibits consistent advantages under zero-shot settings, but is still at a disadvantage compared to fine-tuned models. More deeply, through a series of analytical experiments, we summarize and discuss the challenges faced by LLMs including clustering, domain-specific understanding, and cross-domain in-context learning scenarios. Finally, we provide empirical guidance for future directions to address these challenges.

Large Language Models Meet Open-World Intent Discovery and Recognition: An Evaluation of ChatGPT

**Scoring is only possible in the semifinals hall**

**Abstract:**



Depression and Anxiety - Diagnosis and Management in Primary Care:
Alternate Therapy Modalities: A Review

Background: About half of Minneapolis VA patients reside in rural areas, receiving their primary care at a Community Based Outpatient Clinic (CBOC). Although some CBOC’s are over 200 miles away, patients must travel to the Twin Cities for surgical services. 

Methods: The 167 consecutive patients who opted for telehealth postoperative visits were surveyed. Data collected included travel time and distance to the Minneapolis VA and their local CBOC, need for transportation assistance to the clinic/VA, complications as a result of telehealth and a 1-10 overall satisfaction score. 

Results: Respondents reported a mean + SD satisfaction score of 9.60 ± 1.20, with a mean cost savings of $51.94 ± $40.92, decrease in travel time of 99.4 ± 76.58 minutes and no post-surgical complications missed. 

Conclusions: The telehealth program appears to be safe, saves time and money for veterans and results in extremely high patient satisfaction.


Efficacy of Telehealth for Postoperative Visits at the Minneapolis VA

Introduction-
The internet has become a prominent tool for patients, with an estimated 80% of Americans having turned to the internet for health information. The Literacy Project proports that over half of Americans would struggle to read a book written beyond an 8th grade reading level. Studies have found association between poor patient understanding and worse outcomes and decreased patient satisfaction.

To combat these issues related to the readability of patient education materials, The American Medical Association (AMA) recommends patient health information be written at a 6th grade reading level. Previous studies into the readability of online patient information on a wide variety of topics have revealed much online patient education materials surpass this recommendation. With one recent study of the readability of online patient information on the topic of appendectomies was 12th-14th grade, well beyond the abilities of the average American reader and AMA recommendations. 

Because caring for a surgical wound is a prolonged process and often involves at home care, online resources on this topic may particularly useful to patients. Despite this, investigations into the readability of these sources are lacking. This study provides formal assessment of the current state of the readability of online patient education materials on the topic of surgical wound care.

Methods-
A google search for “surgical wound care” was done with a cleared browser. The first 30 results to meet criteria were analyzed. Advertisements, audiovisual resources, and scientific/professional articles were excluded.

The analysis tool readable.com was used to obtain Flesch Kincaid Grade Level Scores and Gunning Fog Index Scores (measures of grade level required to read) as well as Flesch Reading Ease Scores (measure of general ease of read). Average scores with 95% confidence intervals were calculated to compare to recommended reading levels. Associations between readaility and source were also explored.

Results-
Average Flesch Kincaid Grade Level Score was 6.8 95% CI [6.1, 7.5] and average Gunning Fog Index Score was 8.9 95% CI [8.1, 9.7]. The average Flesch Reading Ease Score was 65.3 95% CI [60.9, 69.7], with scores over 60 indicating overall easy readability. Only 10% (3/30) Flesch Reading Ease Score were above 80. (12/30) met AMA recommendations by Flesch Kincaid grade scores, but none met AMA recommendations by Gunning Fog Index Score. 

Information sources could be categorized as; health information websites, government affiliated organizations, hospitals/health systems, private clinics, and others (ie- insurance company). These categorizations aren’t completely mutually exclusive.

There was no statistical difference in scores between organizations with/without government affiliation. 

Conclusions-
Flesch Kincaid Grade Level Scores indicated nearly half (12/30) met AMA criteria, but Flesch Reading Ease Scores and Gunning Fox Grade Level Scores indicated 3/30 and 0/30 met AMA guidelines, respectively. While this is considerable variance, it resulted from a difference of only 2.1 grade levels between average scores of the different grade level calculation methods. Additionally, half of all materials were still within 2.3 grade levels of AMA recommendations based on Gunning Fox Grade Level Scores. Taken together, this likely indicates that there is a proportion of patient information materials that meet AMA guidelines, but the majority still do not meet AMA recommendations and a significant number of people would likely have some trouble with them. 

While there is inherent complexity to medical topics, there are many simple ways to improve the readability of any writing. These include things like removing unnecessary words and breaking up longer sentences. Given the large number of patient education materials within a couple grade levels of the AMA guideline, simple modifications to writing techniques like these may bring a significant proportion into compliance.


Readability Assessment of Online Patient Education Materials on Surgical Wound Care

Brugada syndrome is characterized by coved-type ST-segment elevations in the precordial leads associated with syncope, life-threatening ventricular tachyarrhythmias and sudden cardiac death. When typical ECG findings are present in the absence of symptoms, it is called Brugada pattern. About 25% of cases are associated with loss-of-function mutation of the cardiac sodium channel SCN5A while others may be triggered by drugs with sodium channel-blocking effects, electrolyte abnormalities, febrile states, or autonomic dysfunction.

A 33-year-old man presented with a 3-day history of intermittent substernal chest pressure. On hospital admission, ECG showed normal sinus rhythm without ST-segment changes and serial cardiac enzymes were normal. Given persistent symptoms and unremarkable recent exercise ECG-stress testing, cardiac CT-angiography was performed which revealed anomalous origin of the right coronary artery (RCA) with approximately 50% luminal narrowing. The patient underwent successful surgical reimplantation of the RCA. Cardiopulmonary bypass was weaned
off and the patient rewarmed from therapeutic hypothermia. Once normothermia was achieved, the patient developed ventricular fibrillation (VF). Following successful resuscitation, ECG revealed ST-segment elevations in V1 and V2 consistent with type 1 Brugada, which completely resolved spontaneously in the subsequent hours and was not reproducible with ajmaline challenge.

It is well known that hyperthermia may trigger Brugada syndrome in predisposed individuals. Studies have shown that mutated sodium channels result in temperature-dependent ionic changes. In febrile states, delayed sodium influx results in heterogeneous depolarization predisposing to VF. On the contrary, severe hypothermia (28.5C) causes sinus bradycardia, QT prolongation, widening of the QRS complex and the pathognomonic Osborn wave. Typical Brugada pattern has been rarely described in hypothermic states. Nonetheless, this pattern has been reproduced in animal models with cooling of the epicardium. Therapeutic hypothermia is targeted in the range of mild hypothermia (32-34C), where VF occurs only rarely, mainly in patients with primary arrhythmic disorders. Tan et al. described a case of transient typical Brugada ECG pattern upon rewarming from therapeutic hypothermia after sudden cardiac death of unknown etiology, with complete resolution with normothermia. These authors proposed that the Brugada pattern was hidden during the hypothermic state. In our patient, however, VF occurred only during the rewarming phase when normothermia was
achieved. We hypothesize that temperature increments, not necessarily hyperthermia, may be associated with increased risk of Brugada syndrome in susceptible patients. Although further research is needed, rewarming from therapeutic hypothermia requires careful monitoring for potentially lethal arrhythmogenic disorders.

A Warming Surprise: Brugada Syndrome After Rewarming from Therapeutic
Hypothermia

Background: Many medical schools have begun including health systems science (HSS) in their curricula to address emerging healthcare needs. With this principle in mind, a clinical reasoning exercise was developed by the Johns Hopkins neurology clerkship team in 2017. Students are given a standardized case of a patient presenting with falls and asked to develop an assessment & plan. During the COVID-19 pandemic, Johns Hopkins replaced tiered grading (honors/high pass/pass/fail) with pass/fail grading for core clerkships. This study seeks to determine if clerkship grading (tiered versus pass/fail) influences the quality of the differential diagnoses, localization of the disease process, clinical reasoning, and diagnostic/management options.

Methods: Data was extracted from the case reports submitted by students in the Neurology clerkship between 2017 and 2022. Report elements in tiered versus pass/fail curricula were compared using chi-squared tests.

Results: Of the 592 students, 306 (51.7%) were in the tiered curriculum, and 286 (48.3%) were in the pass/fail curriculum. With respect to differential diagnoses, students were more likely to include drug-induced parkinsonism (p<0.05) in the tiered system, and Lewy body dementia (p<0.05) in the pass/fail system. The inclusion of other diagnoses was not significantly different between the two systems. For localization, students were more likely to localize to the substantia nigra specifically (p<0.05) in the pass/fail compared to tiered system. Clinical reasoning elements like dysautonomia and timing (p<0.05) were cited more in the tiered system, while discussion of REM sleep behavior was not significantly different between the two systems. Students were more likely to order labs but less likely to order imaging (p<0.05) in the tiered versus pass/fail system. Lastly, for management, students were more likely to request outside consultations/referrals, initiate physical therapy, stop metoclopramide, and treat urinary symptoms and hypotension (p<0.05) in the tiered versus pass/fail system. Treatment with carbidopa-levodopa did not significantly differ between the two systems.

Conclusion: Although clerkship grading does not appear to influence differential formation or clinical reasoning strongly, it does influence diagnostic testing and management choices. Tiered grading correlates with aspects of HSS, including value-based care (deferring imaging, stopping metoclopramide) and interprofessional collaboration (consults, physical therapy).

Looking Backwards to Look Forward: An Analysis of Clinical Reasoning Reports in the Neurology Core Clerkship before and after the transition to Pass/Fail Grading

Background 
Alarm fatigue is a significant issue for clinicians due to excessive exposure to alarms, leading to desensitization and decreased trust in their accuracy. Further, not all alarms in intensive care settings warrant action or immediate attention. This can result in inappropriate alarm silencing, posing risks to patient care, especially in emergencies. To address this, we propose redesigning alarms with descriptive and directional auditory and visual cues, aiming to reduce cognitive load, improve event detection, reaction time, and user satisfaction.

Methods
We designed auditory alarms to differentiate between moderate and severe changes in blood pressure, aiming to improve accuracy and response time by enhancing alarm contrast for varying levels of urgency. Additionally, we developed a custom heads-up display (HUD) to address visual overload from uninformative alarms contributing to alarm fatigue. VUMC providers participated in two trials of auditory testing (Original vs Modified alarms) and visual testing (simulated ICU monitor vs HUD display). After each trial, participants provided subjective feedback on comfort and user satisfaction using NASA-TLX surveys.

Results
Given that p=0.045 for percent correctness in Original vs. Modified alarms, modified auditory alarms improved accuracy, but not response time (p=0.640). However, given that p=0.147 for percent correctness and p=0.323 for response time in First vs. Second trials, accuracy and reaction times did not improve as trials progressed, perhaps suggesting a steep learning curve in familiarization with varying alarm sounds. Interestingly, NASA-TLX survey scores revealed lower average ratings on the metrics of effort and frustration, suggesting improved user comfort with use of modified alarms. 

Preliminary visual data showed no significant difference in percent correct identifications (p = 0.560) with use of HUD. Interestingly, use of HUD trended towards worse performance which may be attributed to lack of familiarity with novel HUD interface. Several participants expressed that they had difficulty readjusting their visual focus between the HUD screen and laptop screen in real time, likely contributing to higher physical demand ratings. 

Conclusions
Our data shows that auditory modifications to alarms are more effective in clinical and intensive care settings compared to traditional alarms. This could significantly improve clinician accuracy in detecting emergent events, leading to better patient outcomes. Future research should explore integrating both auditory and visual modifications to assess their combined impact on accuracy, response time, and user experience. Although user experience was not improved with modified alarms and HUD display, there is potential to enhance the HUD's comfort, wearability, and compatibility with existing ICU monitors.

Mitigating ICU Alarm Fatigue in Clinicians by Integrating Visual and Auditory Feedback

ATPase Inhibitory Factor 1 (IF1) is a nuclear-encoded ATP synthase-interacting mitochondrial protein. IF1 is known to be upregulated during periods of elevated stress, such as during myocardial infarctions or cardiac hypertrophy. Previously conducted lab research has demonstrated that mice with the absence of the protein IF1 were protected against pathological cardiac hypertrophy. To further investigate the extent of such findings, an experiment was designed to investigate the role of the IF1 protein in cardiotoxic conditions induced by Doxorubicin (DOX), a widely used anti-cancer drug. DOX is known to induce cardiotoxicity through the generation of reactive oxygen species, increased levels of lipid peroxidation, and mitochondrial dysfunction. Doxorubicin-induced cardiomyopathy is associated with a low prognosis and can serve to be fatal in certain populations. We hypothesized that mice with the absence of IF1 will demonstrate more preserved cardiac functions in the presence of DOX. 

To test the hypothesis, DOX was intraperitoneally injected into two separate groups of mice: mice for which the IF1 protein was knocked out (IF1-KO) and wildtype mice normally expressing the IF1 protein (IF1-WT). The drug was dissolved in a solvent consisting of dimethyl sulfoxide (DMSO), PEG-300, Tween-80, and distilled water in pre-designated proportions. Injections were performed on three separate days with 6 mg/kg of DOX administered on each injection day. Injections were conducted on Day 0, Day 2, and Day 4. Control mice were injected simply with solvent. Subsequently, transthoracic echocardiograms were performed using a VEVO F2 system at one-week intervals for three weeks ie. Day 7, Day 14, and Day 21. Long-axis M-Mode images were obtained to analyze systolic parameters: ejection fraction, fractional shortening, left ventricular mass, as well as left ventricle anterior and posterior wall thickness. Body mass was also measured initially and after the three-week interval. Findings demonstrated that in the presence of DOX, ejection fraction and fractional shortening, the two parameters of cardiac contraction, significantly decreased in IF1-WT mice; whereas no significant decrease was seen in IF1-KO mice. Furthermore, left ventricular anterior wall thickness was found to be significantly reduced in both IF1-WT and IF1-KO mice, suggesting that DOX may lead to a decrease in heart size. These results demonstrate that the absence of IF1 may provide protection against the cardiotoxic effects of DOX, thus supporting the hypothesis. Further studies could investigate the role of increased doses of DOX on cardiac physiology as well as the effects of increased time intervals through the performance of transthoracic echocardiograms after five to seven weeks. Future experiments using other pathophysiological and biochemical methodologies will gain further insights into the mechanisms of the protective effects of eliminating IF1. Experimental findings may yield new insights into the development of interventions to mitigate adverse effects of this drug in cancer patients under chemotherapy treatments.


Analysis of the Role of ATPase Inhibitory Factor 1 (IF1) in Doxorubicin-Induced Cardiotoxicity

Background: Breast cancer is the second most common cancer amongst women in the United States. From 2013 to 2018, the breast cancer death rate fell by 1%; this decline was attributed to earlier screening and awareness. Early diagnosis of breast cancer can reduce morbidity and the intensity of treatment plans (ex: chemotherapies). There are many different screening tests to predict 5 year and lifetime risk of breast cancer. These screening tests and software programs have large variabilities in their risk prediction. Recent publications show a 5% to 19% overdiagnosis of breast cancer in women. In this study we predicted the 5-year risk, 10-year risk, and lifetime risk of breast cancer for mammogram patients using four different breast cancer screening tools: Mammorisk, Tyer Cuzick 2013 (TC 2013), Tyer Cuzick 2017 (TC 2017), and Gail. We compared their variability using t test statistical analysis. The purpose of our study is to determine the precision of these screening tests and begin streamlining the screening process for breast cancer. 

Methods: We collected Mammorisk data from 242 participants at Southwood Imaging (Youngstown, OH) who came in for their annual mammogram. Most women were between 40 to 70 years old. While participants waited to be screened, we asked them to fill out a questionnaire. The questionnaire contained 19 questions about the patient’s personal health, prior breast biopsies, and family history of breast cancer. We put this additional patient information into the additional screening tests Tyer Cuzick 2013, Tyer Cuzick 2017, and Gail. We calculated the 5-year risk and lifetime risk for each patient. We then measured the 5-year risk, 10-year risk, and lifetime risk assessment using t-test statistical analysis (two tailed). 

Results: The 5-year risk did not significantly differ between breast cancer risk models. The 10-year risk did not significantly differ between breast cancer risk models. The results showed statistically significant differences between lifetime risk for breast cancer risk models. Mammorisk and TC2017 lifetime risks differed by 0.544. The difference in percent lifetime risk between the Mammorisk and TC2013 models was -2.666. Mammorisk and Gail lifetime risks varied by 0.104. Even though our lifetime risk data was significant, the effect size was low . So some of our data may be inconclusive.

Conclusion: Our study indicated a lack of significant difference in 5-year and 10-year risk between models, but lifetime risk did significantly differ between breast cancer risk models. This creates variability that is difficult for physicians to interpret. This can affect physicians’ decision making and future patients’ health outcomes. Additional studies with a larger and more diverse population are needed for definitive evidence in variability between breast cancer risk models. Our study is a first step in finding a streamlined way to approach breast cancer risk in patients. In creating a streamlined approach radiologists will be able to care for cancer patients in a more organized manner.

Variability Between Different Breast Cancer risk Models and Effects on Various Parameters

Acute osteomyelitis in the setting of chronic steroid therapy for the treatment of 17-alpha-hydroxylase deficient congenital adrenal hyperplasia

Lip and face lesions are recognized in dermatology as significantly distressing for patients. There are stereotypes associated with different lesions in this area that affect patients deeply, motivating them to seek care immediately. The average population or clinician is aware of common lip lesions, the most diagnosed of which is herpes simplex virus. Dermatologists must continue to be aware of unique lip and face lesions as they may grow rapidly, be worrisome, and affect the patient’s perception of themselves and psychological well-being. We present the case of neutrophilic sebaceous adenitis, which is an incredibly rare condition that has been reported minimally in the literature. Recognition of rapidly growing lesions, like our patients, is vital for proper diagnosis and treatment.
A 74-year-old female presented to the dermatology clinic with a 5-day history of a rapidly progressive lower lip lesion. Valacyclovir and Carmex only minimally improved her condition. Examination was significant for tenderness to palpation. The patient had no other lesions, no preceding trauma or bites, and no systemic symptoms. Histopathology showed a somewhat dense, superficial and deep, perivascular and periadnexal infiltrate composed predominantly of lymphocytes. The overlying epidermis showed some interface changes with a few necrotic keratinocytes as well as a focal but prominent neutrophilic infiltrate involving a sebaceous gland lobule. No increased dermal mucin was identified on Alcian blue stain. The patient discontinued valacyclovir and treated the lesion with triamcinolone 0.1% twice a day for a few weeks until it cleared. 
Neutrophilic sebaceous adenitis is a very rare condition. Pathogenesis is unknown but triggers such as sun exposure, antibiotics, and febrile illness have been reported. Lesions can progress over days to months and are typically asymptomatic. Recognition of this lesion around the lips and face is imperative as a differential diagnosis for providers to be aware of. Histopathology typically shows an inflammatory infiltrate, particularly of neutrophils and lymphocytes, around sebaceous glands. Sebocytes may be necrosed and further inflammatory infiltrates may be noted around the dermis. There are no current dermatologic guidelines on management although case reports have shown successful treatment with topical or oral steroids, metronidazole, or isotretinoin.

Underline digital video library

Zero-Shot-BERT-Adapters: a Zero-Shot Pipeline for Unknown Intent Detection

Large Language Models Meet Open-World Intent Discovery and Recognition: An Evaluation of ChatGPT

Depression and Anxiety - Diagnosis and Management in Primary Care: Alternate Therapy Modalities: A Review

Efficacy of Telehealth for Postoperative Visits at the Minneapolis VA

Readability Assessment of Online Patient Education Materials on Surgical Wound Care

A Warming Surprise: Brugada Syndrome After Rewarming from Therapeutic Hypothermia

Looking Backwards to Look Forward: An Analysis of Clinical Reasoning Reports in the Neurology Core Clerkship before and after the transition to Pass/Fail Grading

Mitigating ICU Alarm Fatigue in Clinicians by Integrating Visual and Auditory Feedback

Analysis of the Role of ATPase Inhibitory Factor 1 (IF1) in Doxorubicin-Induced Cardiotoxicity

Variability Between Different Breast Cancer risk Models and Effects on Various Parameters

Acute osteomyelitis in the setting of chronic steroid therapy for the treatment of 17-alpha-hydroxylase deficient congenital adrenal hyperplasia

Looking Beyond the Ordinary Lip Lesion: A Rapidly Progressive Case of Neutrophilic Sebaceous Adenitis

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES