Hollow-tree super: A directional and scalable approach for feature importance in boosted tree models

Re-tear after arthroscopic rotator cuff tear surgery: risk analysis using machine learning

BACKGROUND

Postoperative rotator cuff retear after arthroscopic rotator cuff repair (ARCR) is still a major problem. Various risk factors such as age, gender, and tear size have been reported. Recently, magnetic resonance imaging-based stump classification was reported as an index of rotator cuff fragility. Although stump type 3 is reported to have a high retear rate, there are few reports on the risk of postoperative retear based on this classification. Machine learning (ML), an artificial intelligence technique, allows for more flexible predictive models than conventional statistical methods and has been applied to predict clinical outcomes. In this study, we used ML to predict postoperative retear risk after ARCR.

METHODS

The retrospective case-control study included 353 patients who underwent surgical treatment for complete rotator cuff tear using the suture-bridge technique. Patients who initially presented with retears and traumatic tears were excluded. In study participants, after the initial tear repair, rotator cuff retears were diagnosed by magnetic resonance imaging; Sugaya classification types IV and V were defined as re-tears. Age, gender, stump classification, tear size, Goutallier classification, presence of diabetes, and hyperlipidemia were used for ML parameters to predict the risk of retear. Using Python's Scikit-learn as an ML library, five different AI models (logistic regression, random forest, AdaBoost, CatBoost, LightGBM) were trained on the existing data, and the prediction models were applied to the test dataset. The performance of these ML models was measured by the area under the receiver operating characteristic curve. Additionally, key features affecting retear were evaluated.

RESULTS

The area under the receiver operating characteristic curve for logistic regression was 0.78, random forest 0.82, AdaBoost 0.78, CatBoost 0.83, and LightGBM 0.87, respectively for each model. LightGBM showed the highest score. The important factors for model prediction were age, stump classification, and tear size.

CONCLUSIONS

The ML classifier model predicted retears after ARCR with high accuracy, and the AI model showed that the most important characteristics affecting retears were age and imaging findings, including stump classification. This model may be able to predict postoperative rotator cuff retears based on clinical features.

Functional connectivity analysis of the depression connectome provides potential markers and targets for transcranial magnetic stimulation

BACKGROUND

Despite efforts to improve targeting accuracy of the dorsolateral prefrontal cortex (DLPFC) as a repetitive transcranial magnetic stimulation (rTMS) target for Major Depressive Disorder (MDD), the heterogeneity in clinical response remains unexplained.

OBJECTIVE

We sought to compare the patterns of functional connectivity from the DLPFC treatment site in patients with MDD who were TMS responders to those who were TMS non-responders.

METHODS

Baseline anatomical T1 magnetic resonance imaging (MRI), resting-state functional MRI, and diffusion weighted imaging scans were obtained from 37 participants before they underwent a course of rTMS to left Brodmann area 46. A novel machine learning method was utilized to identify brain regions associated with each item of the Beck's Depression Inventory II (BDI-II), and for 26 participants who underwent rTMS treatment over the left Brodmann area 46, identify regions differentiating rTMS responders and non-responders.

RESULTS

Nine parcels of the Human Connectome Project Multimodal Parcellation Atlas matched to at least three items of the Beck's Depression Inventory II (BDI-II) as predictors of response to rTMS, with many in the temporal, parietal and cingulate cortices. Additionally, pre-treatment mapping for 17 items of the BDI-II demonstrated significant variability in symptom to parcel mapping. When parcels associated with symptom presence and symptom resolution were compared, 15 parcels were uniquely associated with resolution (potential targets), and 12 parcels were associated with both symptom presence and resolution (blockers or biomarkers).

CONCLUSIONS

Machine learning approaches show promise for the development of pathoanatomical diagnosis and treatment algorithms for MDD. Prospective studies are required to facilitate clinical translation.

Functional connectivity of the language area in migraine: a preliminary classification model

BACKGROUND

Migraine is a complex disorder characterized by debilitating headaches. Despite its prevalence, its pathophysiology remains unknown, with subsequent gaps in diagnosis and treatment. We combined machine learning with connectivity analysis and applied a whole-brain network approach to identify potential targets for migraine diagnosis and treatment.

METHODS

Baseline anatomical T1 magnetic resonance imaging (MRI), resting-state functional MRI(rfMRI), and diffusion weighted scans were obtained from 31 patients with migraine, and 17 controls. A recently developed machine learning technique, Hollow Tree Super (HoTS) was used to classify subjects into diagnostic groups based on functional connectivity (FC) and derive networks and parcels contributing to the model. PageRank centrality analysis was also performed on the structural connectome to identify changes in hubness.

RESULTS

Our model attained an area under the receiver operating characteristic curve (AUC-ROC) of 0.68, which rose to 0.86 following hyperparameter tuning. FC of the language network was most predictive of the model's classification, though patients with migraine also demonstrated differences in the accessory language, visual and medial temporal regions. Several analogous regions in the right hemisphere demonstrated changes in PageRank centrality, suggesting possible compensation.

CONCLUSIONS

Although our small sample size demands caution, our preliminary findings demonstrate the utility of our method in providing a network-based perspective to diagnosis and treatment of migraine.

Symptom-circuit mappings of the schizophrenia connectome

OBJECTIVE

This paper aims to model the anatomical circuits underlying schizophrenia symptoms, and to explore patterns of abnormal connectivity among brain networks affected by psychopathology.

METHODS

T1 magnetic resonance imaging (MRI), diffusion weighted imaging (DWI), and resting-state functional MRI (rsfMRI) were obtained from a total of 126 patients with schizophrenia who were recruited for the study. The images were processed using the Omniscient software (https://www.o8t. com). We further apply the use of the Hollow-tree Super (HoTS) method to gain insights into what brain regions had abnormal connectivity that might be linked to the symptoms of schizophrenia.

RESULTS

The Positive and Negative Symptom Scale is characterised into 6 factors. Each symptom is mapped with specific anatomical abnormalities and circuits. Comparison between factors reveals co-occurrence in parcels in Factor 1 and Factor 2. Multiple large-scale networks are involved in SCZ symptomatology, with functional connectivity within Default Mode Network (DMN) and Central Executive Network (CEN) regions most frequently associated with measures of psychopathology.

CONCLUSION

We present a summary of the relevant anatomy for regions of the cortical areas as part of a larger effort to understand its contribution in schizophrenia. This unique machine learning-type approach maps symptoms to specific brain regions and circuits by bridging the diagnostic subtypes and analysing the features of the connectome.

Use of machine learning to identify functional connectivity changes in a clinical cohort of patients at risk for dementia

Objective

Progressive conditions characterized by cognitive decline, including mild cognitive impairment (MCI) and subjective cognitive decline (SCD) are clinical conditions representing a major risk factor to develop dementia, however, the diagnosis of these pre-dementia conditions remains a challenge given the heterogeneity in clinical trajectories. Earlier diagnosis requires data-driven approaches for improved and targeted treatment modalities.

Methods

Neuropsychological tests, baseline anatomical T1 magnetic resonance imaging (MRI), resting-state functional MRI (rsfMRI), and diffusion weighted scans were obtained from 35 patients with SCD, 19 with MCI, and 36 age-matched healthy controls (HC). A recently developed machine learning technique, Hollow Tree Super (HoTS) was utilized to classify subjects into diagnostic categories based on their FC, and derive network and parcel-based FC features contributing to each model. The same approach was used to identify features associated with performance in a range of neuropsychological tests. We concluded our analysis by looking at changes in PageRank centrality (a measure of node hubness) between the diagnostic groups.

Results

Subjects were classified into diagnostic categories with a high area under the receiver operating characteristic curve (AUC-ROC), ranging from 0.73 to 0.84. The language networks were most notably associated with classification. Several central networks and sensory brain regions were predictors of poor performance in neuropsychological tests, suggesting maladaptive compensation. PageRank analysis highlighted that basal and limbic deep brain region, along with the frontal operculum demonstrated a reduction in centrality in both SCD and MCI patients compared to controls.

Conclusion

Our methods highlight the potential to explore the underlying neural networks contributing to the cognitive changes and neuroplastic responses in prodromal dementia.

Connectomic disturbances underlying insomnia disorder and predictors of treatment response

Objective

Despite its prevalence, insomnia disorder (ID) remains poorly understood. In this study, we used machine learning to analyze the functional connectivity (FC) disturbances underlying ID, and identify potential predictors of treatment response through recurrent transcranial magnetic stimulation (rTMS) and pharmacotherapy.

Materials and methods

51 adult patients with chronic insomnia and 42 healthy age and education matched controls underwent baseline anatomical T1 magnetic resonance imaging (MRI), resting-stage functional MRI (rsfMRI), and diffusion weighted imaging (DWI). Imaging was repeated for 24 ID patients following four weeks of treatment with pharmacotherapy, with or without rTMS. A recently developed machine learning technique, Hollow Tree Super (HoTS) was used to classify subjects into ID and control groups based on their FC, and derive network and parcel-based FC features contributing to each model. The number of FC anomalies within each network was also compared between responders and non-responders using median absolute deviation at baseline and follow-up.

Results

Subjects were classified into ID and control with an area under the receiver operating characteristic curve (AUC-ROC) of 0.828. Baseline FC anomaly counts were higher in responders than non-responders. Response as measured by the Insomnia Severity Index (ISI) was associated with a decrease in anomaly counts across all networks, while all networks showed an increase in anomaly counts when response was measured using the Pittsburgh Sleep Quality Index. Overall, responders also showed greater change in all networks, with the Default Mode Network demonstrating the greatest change.

Conclusion

Machine learning analysis into the functional connectome in ID may provide useful insight into diagnostic and therapeutic targets.

12 Plagues of AI in Healthcare: A Practical Guide to Current Issues With Using Machine Learning in a Medical Context

The healthcare field has long been promised a number of exciting and powerful applications of Artificial Intelligence (AI) to improve the quality and delivery of health care services. AI techniques, such as machine learning (ML), have proven the ability to model enormous amounts of complex data and biological phenomena in ways only imaginable with human abilities alone. As such, medical professionals, data scientists, and Big Tech companies alike have all invested substantial time, effort, and funding into these technologies with hopes that AI systems will provide rigorous and systematic interpretations of large amounts of data that can be leveraged to augment clinical judgments in real time. However, despite not being newly introduced, AI-based medical devices have more than often been limited in their true clinical impact that was originally promised or that which is likely capable, such as during the current COVID-19 pandemic. There are several common pitfalls for these technologies that if not prospectively managed or adjusted in real-time, will continue to hinder their performance in high stakes environments outside of the lab in which they were created. To address these concerns, we outline and discuss many of the problems that future developers will likely face that contribute to these failures. Specifically, we examine the field under four lenses: approach, data, method and operation. If we continue to prospectively address and manage these concerns with reliable solutions and appropriate system processes in place, then we as a field may further optimize the clinical applicability and adoption of medical based AI technology moving forward.

Machine Learning Decomposition of the Anatomy of Neuropsychological Deficit in Alzheimer’s Disease and Mild Cognitive Impairment

Objective

Alzheimer’s Disease (AD) is a progressive condition characterized by cognitive decline. AD is often preceded by mild cognitive impairment (MCI), though the diagnosis of both conditions remains a challenge. Early diagnosis of AD, and prediction of MCI progression require data-driven approaches to improve patient selection for treatment. We used a machine learning tool to predict performance in neuropsychological tests in AD and MCI based on functional connectivity using a whole-brain connectome, in an attempt to identify network substrates of cognitive deficits in AD.

Methods

Neuropsychological tests, baseline anatomical T1 magnetic resonance imaging (MRI), resting-state functional MRI, and diffusion weighted imaging scans were obtained from 149 MCI, and 85 AD patients; and 140 cognitively unimpaired geriatric participants. A novel machine learning tool, Hollow Tree Super (HoTS) was utilized to extract feature importance from each machine learning model to identify brain regions that were associated with deficit and absence of deficit for 11 neuropsychological tests.

Results

11 models attained an area under the receiver operating curve (AUC-ROC) greater than 0.65, while five models had an AUC-ROC ≥ 0.7. 20 parcels of the Human Connectome Project Multimodal Parcelation Atlas matched to poor performance in at least two neuropsychological tests, while 14 parcels were associated with good performance in at least two tests. At a network level, most parcels predictive of both presence and absence of deficit were affiliated with the Central Executive Network, Default Mode Network, and the Sensorimotor Networks. Segregating predictors by the cognitive domain associated with each test revealed areas of coherent overlap between cognitive domains, with the parcels providing possible markers to screen for cognitive impairment.

Conclusion

Approaches such as ours which incorporate whole-brain functional connectivity and harness feature importance in machine learning models may aid in identifying diagnostic and therapeutic targets in AD.