Towards long-tailed, multi-label disease classification from chest X-ray: Overview of the CXR-LT challenge.

Many real-world image recognition problems, such as diagnostic medical imaging exams, are "long-tailed" - there are a few common findings followed by many more relatively rare conditions. In chest radiography, diagnosis is both a long-tailed and multi-label problem, as patients often present with multiple findings simultaneously. While researchers have begun to study the problem of long-tailed learning in medical image recognition, few have studied the interaction of label imbalance and label co-occurrence posed by long-tailed, multi-label disease classification. To engage with the research community on this emerging topic, we conducted an open challenge, CXR-LT, on long-tailed, multi-label thorax disease classification from chest X-rays (CXRs). We publicly release a large-scale benchmark dataset of over 350,000 CXRs, each labeled with at least one of 26 clinical findings following a long-tailed distribution. We synthesize common themes of top-performing solutions, providing practical recommendations for long-tailed, multi-label medical image classification. Finally, we use these insights to propose a path forward involving vision-language foundation models for few- and zero-shot disease classification.

Does Elevated Mean-Arterial Pressure Lead to Better Outcomes in Degenerative Cervical Myelopathy?- A Prospective, Pilot Randomized Control Trial.

STUDY DESIGN: Randomized Control Trial. OBJECTIVE: DCM refers to compression of spinal cord either due to static/dynamic causes or commonly, a result of combination of both. Number of variables exist, which determine prognosis post-surgery. Role of intra-operative blood pressure has not been analyzed in depth in current literature. Elevating MAP post SCI is widely practiced and forms a recommendation of AANS/CNS Joint Committee Guidelines. This led us to investigate role played by elevated MAP during surgery for DCM, in order to optimize outcomes. METHODS: This prospective randomized comparative pilot study was conducted at a tertiary care spine centre. 84 patients were randomly divided in two groups. Group 1 had intra-operative MAP in normal range. Group 2, had intra-operative BP 20 mmHg higher than preoperative average MAP with a variation of + 5 mmHg. Outcomes were recorded at 3 months, 6 months and 1 year by mJOA, VAS and ASIA scale. RESULTS: Neurological improvement was documented in 19/30 (63.3%) patients of hypertensive group compared to 16/30 (53.3%) patients of normotensive group. Improvements in mJOA scores were better for hypertensive group during the 1-year follow-up. Improvement in VAS scores were comparable between two groups, but at 1-year follow-up the VAS score of hypertensive groups was significantly lower. CONCLUSION: MAP should be individualized according to preoperative average blood pressure assessment of patient. Keeping intraoperative MAP at higher level (preoperative MAP + 20 mmHg) during surgery for DCM can result in better outcomes.

CancerGPT for few shot drug pair synergy prediction using large pretrained language models.

Large language models (LLMs) have been shown to have significant potential in few-shot learning across various fields, even with minimal training data. However, their ability to generalize to unseen tasks in more complex fields, such as biology and medicine has yet to be fully evaluated. LLMs can offer a promising alternative approach for biological inference, particularly in cases where structured data and sample size are limited, by extracting prior knowledge from text corpora. Here we report our proposed few-shot learning approach, which uses LLMs to predict the synergy of drug pairs in rare tissues that lack structured data and features. Our experiments, which involved seven rare tissues from different cancer types, demonstrate that the LLM-based prediction model achieves significant accuracy with very few or zero samples. Our proposed model, the CancerGPT (with ~ 124M parameters), is comparable to the larger fine-tuned GPT-3 model (with ~ 175B parameters). Our research contributes to tackling drug pair synergy prediction in rare tissues with limited data, and also advancing the use of LLMs for biological and medical inference tasks.

Towards long-tailed, multi-label disease classification from chest X-ray: Overview of the CXR-LT challenge.

Many real-world image recognition problems, such as diagnostic medical imaging exams, are "long-tailed" - there are a few common findings followed by many more relatively rare conditions. In chest radiography, diagnosis is both a long-tailed and multi-label problem, as patients often present with multiple findings simultaneously. While researchers have begun to study the problem of long-tailed learning in medical image recognition, few have studied the interaction of label imbalance and label co-occurrence posed by long-tailed, multi-label disease classification. To engage with the research community on this emerging topic, we conducted an open challenge, CXR-LT, on long-tailed, multi-label thorax disease classification from chest X-rays (CXRs). We publicly release a large-scale benchmark dataset of over 350,000 CXRs, each labeled with at least one of 26 clinical findings following a long-tailed distribution. We synthesize common themes of top-performing solutions, providing practical recommendations for long-tailed, multi-label medical image classification. Finally, we use these insights to propose a path forward involving vision-language foundation models for few- and zero-shot disease classification.

How Does Pruning Impact Long-Tailed Multi-Label Medical Image Classifiers?

Pruning has emerged as a powerful technique for compressing deep neural networks, reducing memory usage and inference time without significantly affecting overall performance. However, the nuanced ways in which pruning impacts model behavior are not well understood, particularly for long-tailed, multi-label datasets commonly found in clinical settings. This knowledge gap could have dangerous implications when deploying a pruned model for diagnosis, where unexpected model behavior could impact patient well-being. To fill this gap, we perform the first analysis of pruning's effect on neural networks trained to diagnose thorax diseases from chest X-rays (CXRs). On two large CXR datasets, we examine which diseases are most affected by pruning and characterize class "forgettability" based on disease frequency and co-occurrence behavior. Further, we identify individual CXRs where uncompressed and heavily pruned models disagree, known as pruning-identified exemplars (PIEs), and conduct a human reader study to evaluate their unifying qualities. We find that radiologists perceive PIEs as having more label noise, lower image quality, and higher diagnosis difficulty. This work represents a first step toward understanding the impact of pruning on model behavior in deep long-tailed, multi-label medical image classification. All code, model weights, and data access instructions can be found at https://github.com/VITA-Group/PruneCXR.

How Does Pruning Impact Long-Tailed Multi-label Medical Image Classifiers?

Pruning has emerged as a powerful technique for compressing deep neural networks, reducing memory usage and inference time without significantly affecting overall performance. However, the nuanced ways in which pruning impacts model behavior are not well understood, particularly for long-tailed, multi-label datasets commonly found in clinical settings. This knowledge gap could have dangerous implications when deploying a pruned model for diagnosis, where unexpected model behavior could impact patient well-being. To fill this gap, we perform the first analysis of pruning's effect on neural networks trained to diagnose thorax diseases from chest X-rays (CXRs). On two large CXR datasets, we examine which diseases are most affected by pruning and characterize class "forgettability" based on disease frequency and co-occurrence behavior. Further, we identify individual CXRs where uncompressed and heavily pruned models disagree, known as pruning-identified exemplars (PIEs), and conduct a human reader study to evaluate their unifying qualities. We find that radiologists perceive PIEs as having more label noise, lower image quality, and higher diagnosis difficulty. This work represents a first step toward understanding the impact of pruning on model behavior in deep long-tailed, multi-label medical image classification. All code, model weights, and data access instructions can be found at https://github.com/VITA-Group/PruneCXR.

CancerGPT: Few-shot Drug Pair Synergy Prediction using Large Pre-trained Language Models.

Large pre-trained language models (LLMs) have been shown to have significant potential in few-shot learning across various fields, even with minimal training data. However, their ability to generalize to unseen tasks in more complex fields, such as biology, has yet to be fully evaluated. LLMs can offer a promising alternative approach for biological inference, particularly in cases where structured data and sample size are limited, by extracting prior knowledge from text corpora. Our proposed few-shot learning approach uses LLMs to predict the synergy of drug pairs in rare tissues that lack structured data and features. Our experiments, which involved seven rare tissues from different cancer types, demonstrated that the LLM-based prediction model achieved significant accuracy with very few or zero samples. Our proposed model, the CancerGPT (with ~ 124M parameters), was even comparable to the larger fine-tuned GPT-3 model (with ~ 175B parameters). Our research is the first to tackle drug pair synergy prediction in rare tissues with limited data. We are also the first to utilize an LLM-based prediction model for biological reaction prediction tasks.

Attend Who is Weak: Pruning-assisted Medical Image Localization under Sophisticated and Implicit Imbalances.

Deep neural networks (DNNs) have rapidly become a de facto choice for medical image understanding tasks. However, DNNs are notoriously fragile to the class imbalance in image classification. We further point out that such imbalance fragility can be amplified when it comes to more sophisticated tasks such as pathology localization, as imbalances in such problems can have highly complex and often implicit forms of presence. For example, different pathology can have different sizes or colors (w.r.t.the background), different underlying demographic distributions, and in general different difficulty levels to recognize, even in a meticulously curated balanced distribution of training data. In this paper, we propose to use pruning to automatically and adaptively identify hard-to-learn (HTL) training samples, and improve pathology localization by attending them explicitly, during training in supervised, semi-supervised, and weakly-supervised settings. Our main inspiration is drawn from the recent finding that deep classification models have difficult-to-memorize samples and those may be effectively exposed through network pruning [15] - and we extend such observation beyond classification for the first time. We also present an interesting demographic analysis which illustrates HTLs ability to capture complex demographic imbalances. Our extensive experiments on the Skin Lesion Localization task in multiple training settings by paying additional attention to HTLs show significant improvement of localization performance by ~2-3%.

Further divided gender gaps in research productivity and collaboration during the COVID-19 pandemic: Evidence from coronavirus-related literature.

Based on publication data on coronavirus-related fields, this study applies a difference in differences approach to explore the evolution of gender inequalities before and during the COVID-19 pandemic by comparing the differences in the numbers and shares of authorships, leadership in publications, gender composition of collaboration, and scientific impacts. We find that, during the pandemic: (1) females' leadership in publications as the first author was negatively affected; (2) although both females and males published more papers relative to the pre-pandemic period, the gender gaps in the share of authorships have been strengthened due to the larger increase in males' authorships; (3) the share of publications by mixed-gender collaboration declined; (4) papers by teams in which females play a key role were less cited in the pre-pandemic period, and this citation disadvantage was exacerbated during the pandemic; and (5) gender inequalities regarding authorships and collaboration were enhanced in the initial stage of COVID-19, widened with the increasing severity of COVID-19, and returned to the pre-pandemic level in September 2020. This study shows that females' lower participation in teams as major contributors and less collaboration with their male colleagues also reflect their underrepresentation in science in the pandemic period. This investigation significantly deepens our understanding of how the pandemic influenced academia, based on which science policies and gender policy changes are proposed to mitigate the gender gaps.

A Mid-term Comparison of the Functional Outcomes of Medial Pivot and Rotating Platform Mobile-Bearing Total Knee Arthroplasty in the Indian Population.

BACKGROUND: Both medial pivot (MP) and rotating platform (RP) mobile-bearing (MB) total knee arthroplasty (TKA) have been developed to better mimic the natural knee kinematics and femoral roll back in flexion. The purpose of this retrospective study was to compare the mid-term functional outcomes and range of motion (ROM) of MP and RP types of total knee arthroplasty. METHODS: 116 patients (mean age of 66.3 years) undergoing TKA (52 Medial pivot design and 64 Rotating Platform design) were evaluated retrospectively with Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) knee score, knee society score (KSS) with its subgroups namely, Knee Score (KSKS) and Functional Score (KSFS) and forgotten joint score (FJS) at a mean follow-up of 7.1 years. Range of motion (ROM) and tibiofemoral anatomic angle on the radiographs were also compared. RESULTS: Mean ROM, WOMAC and KSKS improved significantly from pre-operative to postoperative knees in both the groups. There was, however, no significant difference between the two groups at the final follow-up. In contrast, mean KSFS score improved to 89.5 ± 8.1 in MP group and 86.3 ± 7.1 in RP Group (p = 0.025), while mean FJS was 85.6 ± 4.1 and 80.9 ± 5.4 in the MP and RP groups, respectively (p = < 0.0001). CONCLUSION: Satisfactory clinical and functional outcomes can be obtained using either a MP or RP knee joint in tricompartmental osteoarthritis of knee. The MP design scores better on the KSFS score and FJS than the RP-TKA.

RoS-KD: A Robust Stochastic Knowledge Distillation Approach for Noisy Medical Imaging.

AI-powered Medical Imaging has recently achieved enormous attention due to its ability to provide fast-paced healthcare diagnoses. However, it usually suffers from a lack of high-quality datasets due to high annotation cost, inter-observer variability, human annotator error, and errors in computer-generated labels. Deep learning models trained on noisy labelled datasets are sensitive to the noise type and lead to less generalization on the unseen samples. To address this challenge, we propose a Robust Stochastic Knowledge Distillation (RoS-KD) framework which mimics the notion of learning a topic from multiple sources to ensure deterrence in learning noisy information. More specifically, RoS-KD learns a smooth, well-informed, and robust student manifold by distilling knowledge from multiple teachers trained on overlapping subsets of training data. Our extensive experiments on popular medical imaging classification tasks (cardiopulmonary disease and lesion classification) using real-world datasets, show the performance benefit of RoS-KD, its ability to distill knowledge from many popular large networks (ResNet-50, DenseNet-121, MobileNet-V2) in a comparatively small network, and its robustness to adversarial attacks (PGD, FSGM). More specifically, RoS-KD achieves > 2% and > 4% improvement on F1-score for lesion classification and cardiopulmonary disease classification tasks, respectively, when the underlying student is ResNet-18 against recent competitive knowledge distillation baseline. Additionally, on cardiopulmonary disease classification task, RoS-KD outperforms most of the SOTA baselines by ~1% gain in AUC score.

Prospective Randomized Control Pilot Study to Compare the Role of Injection Cerebrolysin in Operated cases of Degenerative Cervical Myelopathy.

STUDY DESIGN: Prospective randomized control trial. OBJECTIVE: The aim of this study was to analyze role of cerebrolysin in patients of degenerative cervical myelopathy (DCM) managed by surgical modalities. SUMMARY OF BACKGROUND DATA: Cerebrolysin has been extensively researched with variable success in neurodegenerative pathologies. There has been only one study in published literature till date that has studied role of cerebrolysin in DCM in conservatively managed patients but none in the patients treated surgically. We present our pilot study which analyzes the role of cerebrolysin in patients of DCM managed by surgical modalities. METHODS: This prospective randomized control trial was conducted at a tertiary care institute in Mumbai. Sixty operated cases of DCM were randomly divided into 2 groups. The first group was given Injection Cerebrolysin 5 mL diluted in 100 mL Normal Saline over 30 minutes once a day for 21 days postoperatively. The second group was given placebo. Modified Japanese Orthopedic Association scores (mJOA) and visual analog scale (VAS) were used to document functional outcomes at 3 weeks, 3 months, 6 months, and 1 year. Recovery of hand function was separately accessed by improvement in hand power and sensations. RESULTS: Preoperative mJOA and VAS scores were comparable between 2 groups. Both groups showed significant improvement in both mJOA and VAS scores at 3weeks, 3 months, 6 months and 1-year follow-up (P < 0.01). In comparing the two groups, there was no difference in improvement of mJOA and VAS scores. However, cerebrolysin group showed significant improvement in hand function at 1 year compared to the placebo. Postoperative neurological recovery was better in the cerebrolysin group with 66.7% patients showing complete neurological recovery compared to 56.7% for placebo, but this was statistically insignificant. Two patients developed headache and one patient complained of dizziness in the cerebrolysin group, but these resolved without any intervention. CONCLUSION: Use of cerebrolysin in postoperative cases of DCM is safe and results in improved hand function.Level of Evidence: 1.

Incidence of Dural Tears in Open versus Minimally Invasive Spine Surgery: A Single-Center Prospective Study.

STUDY DESIGN: A prospective comparative study. PURPOSE: To compare the incidence of unintended durotomy and return to work after open surgery versus minimally invasive spine surgery (MIS) for degenerative lumbar pathologies. OVERVIEW OF LITERATURE: The incidence of accidental durotomy varies between 0.3% and 35%. Most of these are from open surgeries, and only a handful of studies have involved the MIS approach. No single-center studies have compared open surgery with MIS, especially in the context of early return to work and dural tear (DT). METHODS: This study included 420 operated cases of degenerative lumbar pathology with a prospective follow-up of at least 6 months. Patients were divided into the open surgery and MIS groups, and the incidences of DT, early return to work, and various demographic and operative factors were compared. RESULTS: A total of 156 and 264 patients underwent MIS and open surgery, respectively. Incidental durotomy was documented in 52 cases (12.4%); this was significantly less in the MIS group versus the open surgery group (6.4% vs. 15.9%, p <0.05). In the open surgery group, four patients underwent revision for persistent dural leak or pseudomeningocele, but none of the cases in the MIS group had revision surgery due to DT-related complications. The incidence of DT was higher among patients with high body mass index, patients with diabetes mellitus, and patients who underwent revision surgery (p <0.05) regardless of the approach. The MIS group returned to work significantly earlier. CONCLUSIONS: MIS was associated with a significantly lower incidence of DT and earlier return to work compared with open surgery among patients with degenerative lumbar pathology.

Does a high BMI affect the outcome of minimally invasive TLIF? A retrospective study of 207 patients.

PURPOSE: We investigated whether a high Body Mass Index (BMI) affects the outcomes following Minimally Invasive TLIF (MI-TLIF) for degenerative lumbar pathologies. METHODS: A retrospective study was undertaken to include patients operated between January 2016 and January 2020 with at least one-year follow-up. Various preoperative and demographic parameters were recorded and the patients were classified into normal, overweight and obese based on the BMI. The operative and outcome measures used for assessment were surgical time, blood loss, number of levels operated upon, skin incision length, day of independent mobilisation, total hospital stay including ICU stay, return to work and Visual Analogue Score (VAS) for back pain (VAS-BP) and leg pain (VAS-LP) and Oswestry Disability Index (ODI). Attainment of Minimal Clinically Important Difference (MCID) for the scores was calculated. Multivariate analyses were done to assess the effect of BMI on different parameters. RESULTS: Blood loss and postoperative ICU stay were found to be higher in the obese patients. However, the other variables were comparable. VAS-BP, VAS-LP and ODI scores were significantly improved in all the patients with no inter-group variability. The MCID attainment was also similar. The satisfaction rating at 1-year and willingness for surgery again for similar disease was also similar. The overall complication rate was 14.9% and was comparable among the groups. Multivariate analyses revealed no significant association between BMI and various parameters. CONCLUSION: In patients treated by MI-TLIF for degenerative lumbar spine pathology, BMI is not a factor that negatively affects the functional and clinical outcomes.

RadBERT-CL: Factually-Aware Contrastive Learning For Radiology Report Classification.

Radiology reports are unstructured and contain the imaging findings and corresponding diagnoses transcribed by radiologists which include clinical facts and negated and/or uncertain statements. Extracting pathologic findings and diagnoses from radiology reports is important for quality control, population health, and monitoring of disease progress. Existing works, primarily rely either on rule-based systems or transformer-based pre-trained model fine-tuning, but could not take the factual and uncertain information into consideration, and therefore generate false positive outputs. In this work, we introduce three sedulous augmentation techniques which retain factual and critical information while generating augmentations for contrastive learning. We introduce RadBERT-CL, which fuses these information into BlueBert via a self-supervised contrastive loss. Our experiments on MIMIC-CXR show superior performance of RadBERT-CL on fine-tuning for multi-class, multi-label report classification. We illustrate that when few labeled data are available, RadBERT-CL outperforms conventional SOTA transformers (BERT/BlueBert) by significantly larger margins (6-11%). We also show that the representations learned by RadBERT-CL can capture critical medical information in the latent space.

SCALP - Supervised Contrastive Learning for Cardiopulmonary Disease Classification and Localization in Chest X-rays using Patient Metadata.

Computer-aided diagnosis plays a salient role in more accessible and accurate cardiopulmonary diseases classification and localization on chest radiography. Millions of people get affected and die due to these diseases without an accurate and timely diagnosis. Recently proposed contrastive learning heavily relies on data augmentation, especially positive data augmentation. However, generating clinically-accurate data augmentations for medical images is extremely difficult because the common data augmentation methods in computer vision, such as sharp, blur, and crop operations, can severely alter the clinical settings of medical images. In this paper, we proposed a novel and simple data augmentation method based on patient metadata and supervised knowledge to create clinically accurate positive and negative augmentations for chest X-rays. We introduce an end-to-end framework, SCALP, which extends the self-supervised contrastive approach to a supervised setting. Specifically, SCALP pulls together chest X-rays from the same patient (positive keys) and pushes apart chest X-rays from different patients (negative keys). In addition, it uses ResNet-50 along with the triplet-attention mechanism to identify cardiopulmonary diseases, and Grad-CAM++ to highlight the abnormal regions. Our extensive experiments demonstrate that SCALP outperforms existing baselines with significant margins in both classification and localization tasks. Specifically, the average classification AUCs improve from 82.8% (SOTA using DenseNet-121) to 83.9% (SCALP using ResNet-50), while the localization results improve on average by 3.7% over different IoU thresholds.

Using Radiomics as Prior Knowledge for Thorax Disease Classification and Localization in Chest X-rays.

Chest X-ray becomes one of the most common medical diagnoses due to its noninvasiveness. The number of chest X-ray images has skyrocketed, but reading chest X-rays still have been manually performed by radiologists, which creates huge burnouts and delays. Traditionally, radiomics, as a subfield of radiology that can extract a large number of quantitative features from medical images, demonstrates its potential to facilitate medical imaging diagnosis before the deep learning era. In this paper, we develop an end-to-end framework, ChexRadiNet, that can utilize the radiomics features to improve the abnormality classification performance. Specifically, ChexRadiNet first applies a light-weight but efficient triplet-attention mechanism to classify the chest X-rays and highlight the abnormal regions. Then it uses the generated class activation map to extract radiomic features, which further guides our model to learn more robust image features. After a number of iterations and with the help of radiomic features, our framework can converge to more accurate image regions. We evaluate the ChexRadiNet framework using three public datasets: NIH ChestX-ray, CheXpert, and MIMIC-CXR. We find that ChexRadiNet outperforms the state-of-the-art on both disease detection (0.843 in AUC) and localization (0.679 in T(IoU) = 0.1). We make the code publicly available at https://github. com/bionlplab/lung_disease_detection_amia2021, with the hope that this method can facilitate the development of automatic systems with a higher-level understanding of the radiological world.

Preoperative T1 magnetic resonance imaging changes carry a poor postoperative prognosis in cervical myelopathy: A retrospective study of 182 patients.

Background: T2 scans are widely used to determine the prognosis for patients undergoing surgery for cervical myelopathy. In this study, we determined whether T1 MR changes in addition to T2 MR changes could have prognostic importance. Methods: This retrospective analysis involved 182 patients undergoing surgery for cervical myelopathy (2017-2020). There were 110 patients in Group 1 (only T2 MR changes) and 72 in Group 2 (both T1 and T2 MR changes). In addition, demographic, visual analog score (VAS), modified Japanese Orthopaedic Association (mJOA) scores, and operative details were recorded at 1 month, 3 months, 6 months, and 1 year postoperatively. Results: Notably, VAS scores were comparable at each point in time and were significantly better than the preoperative scores at 1 year postoperatively. Although mJOA scores were comparable at 1 month in both groups, they were better thereafter for Group 1 patients. Conclusion: The presence of T1 changes on the preoperative magnetic resonance imaging represented a poor prognostic indicator for the postoperative outcome compared to the presence of T2 changes alone.

Conjoint Nerve Root an Intraoperative Challenge in Minimally Invasive Tubular Discectomy.

Conjoint nerve root (CNR) is an embryological nerve root anomaly that mainly involves the lumbosacral region. The presence of CNR during tubular discectomy raises the chances of failure in spinal surgery and the risk of neural injuries. Tubular discectomy can be challenging in the presence of CNR owing to limited visualization. Here, we present a technical note on two cases of L5-S1 disc prolapse in the presence of conjoint S1 nerve root that was operated via a minimally invasive tubular approach. Any intraoperative suspicion of CNR while using the tubular approach should prompt the surgeon to perform a thorough tubular decompression prior to nerve root retraction. In patients with a large disc, disc should be approached via the axilla because the axillary area between the dura and the medial boarder of the root is very easy to approach in the presence of CNR. Safe performance of tubular discectomy is possible even in the presence of CNR in the lumbar spine.