Harnessing artificial intelligence in radiology to augment population health

This review article serves to highlight radiological services as a major cost driver for the healthcare sector, and the potential improvements in productivity and cost savings that can be generated by incorporating artificial intelligence (AI) into the radiology workflow, referencing Singapore healthcare as an example. More specifically, we will discuss the opportunities for AI in lowering healthcare costs and supporting transformational shifts in our care model in the following domains: predictive analytics for optimising throughput and appropriate referrals, computer vision for image enhancement (to increase scanner efficiency and decrease radiation exposure) and pattern recognition (to aid human interpretation and worklist prioritisation), natural language processing and large language models for optimising reports and text data-mining. In the context of preventive health, we will discuss how AI can support population level screening for major disease burdens through opportunistic screening and democratise expertise to increase access to radiological services in primary and community care.

Neuroanatomical subtypes of schizophrenia and relationship with illness duration and deficit status

BACKGROUND

The heterogeneity of schizophrenia (SCZ) regarding psychopathology, illness trajectory and their inter-relationships with underlying neural substrates remain incompletely understood. In a bid to reduce illness heterogeneity using neural substrates, our study aimed to replicate the findings of an earlier study by Chand et al. (2020). We employed brain structural measures for subtyping SCZ patients, and evaluate each subtype's relationship with clinical features such as illness duration, psychotic psychopathology, and additionally deficit status.

METHODS

Overall, 240 subjects (160 SCZ patients, 80 healthy controls) were recruited for this study. The participants underwent brain structural magnetic resonance imaging scans and clinical rating using the Positive and Negative Syndrome Scale. Neuroanatomical subtypes of SCZ were identified using "Heterogeneity through discriminative analysis" (HYDRA), a clustering technique which accounted for relevant covariates and the inter-group normalized percentage changes in brain volume were also calculated.

RESULTS

As replicated, two neuroanatomical subtypes (SG-1 and SG-2) were found amongst our patients with SCZ. The subtype SG-1 was associated with enlargements in the third and lateral ventricles, volume increase in the basal ganglia (putamen, caudate, pallidum), longer illness duration, and deficit status. The subtype SG-2 was associated with reductions of cortical and subcortical structures (hippocampus, thalamus, basal ganglia).

CONCLUSIONS

These replicated findings have clinical implications in the early intervention, response monitoring, and prognostication of SCZ. Future studies may adopt a multi-modal neuroimaging approach to enhance insights into the neurobiological composition of relevant subtypes.

Machine learning classification of schizophrenia patients and healthy controls using diverse neuroanatomical markers and Ensemble methods

Schizophrenia is a major psychiatric disorder that imposes enormous clinical burden on patients and their caregivers. Determining classification biomarkers can complement clinical measures and improve understanding of the neural basis underlying schizophrenia. Using neuroanatomical features, several machine learning based investigations have attempted to classify schizophrenia from healthy controls but the range of neuroanatomical measures employed have been limited in range to date. In this study, we sought to classify schizophrenia and healthy control cohorts using a diverse set of neuroanatomical measures (cortical and subcortical volumes, cortical areas and thickness, cortical mean curvature) and adopted Ensemble methods for better performance. Additionally, we correlated such neuroanatomical features with Quality of Life (QoL) assessment scores within the schizophrenia cohort. With Ensemble methods and diverse neuroanatomical measures, we achieved classification accuracies ranging from 83 to 87%, sensitivities and specificities varying between 90-98% and 65-70% respectively. In addition to lower QoL scores within schizophrenia cohort, significant correlations were found between specific neuroanatomical measures and psychological health, social relationship subscale domains of QoL. Our results suggest the utility of inclusion of subcortical and cortical measures and Ensemble methods to achieve better classification performance and their potential impact of parsing out neurobiological correlates of quality of life in schizophrenia.

Early postnatal irradiation-induced age-dependent changes in adult mouse brain: MRI based characterization

Background

Brain radiation exposure, in particular, radiotherapy, can induce cognitive impairment in patients, with significant effects persisting for the rest of their life. However, the main mechanisms leading to this adverse event remain largely unknown. A study of radiation-induced injury to multiple brain regions, focused on the hippocampus, may shed light on neuroanatomic bases of neurocognitive impairments in patients. Hence, we irradiated BALB/c mice (male and female) at postnatal day 3 (P3), day 10 (P10), and day 21 (P21) and investigated the long-term radiation effect on brain MRI changes and hippocampal neurogenesis.

Results

We found characteristic brain volume reductions in the hippocampus, olfactory bulbs, the cerebellar hemisphere, cerebellar white matter (WM) and cerebellar vermis WM, cingulate, occipital and frontal cortices, cerebellar flocculonodular WM, parietal region, endopiriform claustrum, and entorhinal cortex after irradiation with 5 Gy at P3. Irradiation at P10 induced significant volume reduction in the cerebellum, parietal region, cingulate region, and olfactory bulbs, whereas the reduction of the volume in the entorhinal, parietal, insular, and frontal cortices was demonstrated after irradiation at P21. Immunohistochemical study with cell division marker Ki67 and immature marker doublecortin (DCX) indicated the reduced cell division and genesis of new neurons in the subgranular zone of the dentate gyrus in the hippocampus after irradiation at all three postnatal days, but the reduction of total granule cells in the stratum granulosun was found after irradiation at P3 and P10.

Conclusions

The early life radiation exposure during different developmental stages induces varied brain pathophysiological changes which may be related to the development of neurological and neuropsychological disorders later in life.

CD10 marks non-canonical PPARγ-independent adipocyte maturation and browning potential of adipose-derived stem cells

BACKGROUND

Effective stem cell therapy is dependent on the stem cell quality that is determined by their differentiation potential, impairment of which leads to poor engraftment and survival into the target cells. However, limitations in our understanding and the lack of reliable markers that can predict their maturation efficacies have hindered the development of stem cells as an effective therapeutic strategy. Our previous study identified CD10, a pro-adipogenic, depot-specific prospective cell surface marker of human adipose-derived stem cells (ASCs). Here, we aim to determine if CD10 can be used as a prospective marker to predict mature adipocyte quality and play a direct role in adipocyte maturation.

METHODS

We first generated 14 primary human subject-derived ASCs and stable immortalized CD10 knockdown and overexpression lines for 4 subjects by the lentiviral transduction system. To evaluate the role of CD10 in adipogenesis, the adipogenic potential of the human subject samples were scored against their respective CD10 transcript levels. Assessment of UCP1 expression levels was performed to correlate CD10 levels to the browning potential of mature ASCs. Quantitative polymerase chain reaction (qPCR) and Western blot analysis were performed to determine CD10-dependent regulation of various targets. Seahorse analysis of oxidative metabolism and lipolysis assay were studied. Lastly, as a proof-of-concept study, we used CD10 as a prospective marker for screening nuclear receptor ligands library.

RESULTS

We identified intrinsic CD10 levels as a positive determinant of adipocyte maturation as well as browning potential of ASCs. Interestingly, CD10 regulates ASC's adipogenic maturation non-canonically by modulating endogenous lipolysis without affecting the classical peroxisome proliferator-activated receptor gamma (PPARγ)-dependent adipogenic pathways. Furthermore, our CD10-mediated screening analysis identified dexamethasone and retinoic acid as stimulator and inhibitor of adipogenesis, respectively, indicating CD10 as a useful biomarker for pro-adipogenic drug screening.

CONCLUSION

Overall, we establish CD10 as a functionally relevant ASC biomarker, which may be a prerequisite to identify high-quality cell populations for improving metabolic diseases.

Fast Adipogenesis Tracking System (FATS)-a robust, high-throughput, automation-ready adipogenesis quantification technique

Adipogenesis is essential in in vitro experimentation to assess differentiation capability of stem cells, and therefore, its accurate measurement is important. Quantitative analysis of adipogenic levels, however, is challenging and often susceptible to errors due to non-specific reading or manual estimation by observers. To this end, we developed a novel adipocyte quantification algorithm, named Fast Adipogenesis Tracking System (FATS), based on computer vision libraries. The FATS algorithm is versatile and capable of accurately detecting and quantifying percentage of cells undergoing adipogenic and browning differentiation even under difficult conditions such as the presence of large cell clumps or high cell densities. The algorithm was tested on various cell lines including 3T3-L1 cells, adipose-derived mesenchymal stem cells (ASCs), and induced pluripotent stem cell (iPSC)-derived cells. The FATS algorithm is particularly useful for adipogenic measurement of embryoid bodies derived from pluripotent stem cells and was capable of accurately distinguishing adipogenic cells from false-positive stains. We then demonstrate the effectiveness of the FATS algorithm for screening of nuclear receptor ligands that affect adipogenesis in the high-throughput manner. Together, the FATS offer a universal and automated image-based method to quantify adipocyte differentiation of different cell lines in both standard and high-throughput workflows.

Adipocyte Ceramides Regulate Subcutaneous Adipose Browning, Inflammation, and Metabolism

Adipocytes package incoming fatty acids into triglycerides and other glycerolipids, with only a fraction spilling into a parallel biosynthetic pathway that produces sphingolipids. Herein, we demonstrate that subcutaneous adipose tissue of type 2 diabetics contains considerably more sphingolipids than non-diabetic, BMI-matched counterparts. Whole-body and adipose tissue-specific inhibition/deletion of serine palmitoyltransferase (Sptlc), the first enzyme in the sphingolipid biosynthesis cascade, in mice markedly altered adipose morphology and metabolism, particularly in subcutaneous adipose tissue. The reduction in adipose sphingolipids increased brown and beige/brite adipocyte numbers, mitochondrial activity, and insulin sensitivity. The manipulation also increased numbers of anti-inflammatory M2 macrophages in the adipose bed and induced secretion of insulin-sensitizing adipokines. By comparison, deletion of serine palmitoyltransferase from macrophages had no discernible effects on metabolic homeostasis or adipose function. These data indicate that newly synthesized adipocyte sphingolipids are nutrient signals that drive changes in the adipose phenotype to influence whole-body energy expenditure and nutrient metabolism.

Effect of Exercise and Calorie Restriction on Tissue Acylcarnitines, Tissue Desaturase Indices, and Fat Accumulation in Diet-Induced Obese Rats

Both exercise and calorie restriction interventions have been recommended for inducing weight-loss in obese states. However, there is conflicting evidence on their relative benefits for metabolic health and insulin sensitivity. This study seeks to evaluate the differential effects of the two interventions on fat mobilization, fat metabolism, and insulin sensitivity in diet-induced obese animal models. After 4 months of ad libitum high fat diet feeding, 35 male Fischer F344 rats were grouped (n = 7 per cohort) into sedentary control (CON), exercise once a day (EX1), exercise twice a day (EX2), 15% calorie restriction (CR1) and 30% calorie restriction (CR2) cohorts. Interventions were carried out over a 4-week period. We found elevated hepatic and muscle long chain acylcarnitines with both exercise and calorie restriction, and a positive association between hepatic long chain acylcarnitines and insulin sensitivity in the pooled cohort. Our result suggests that long chain acylcarnitines may not indicate incomplete fat oxidation in weight loss interventions. Calorie restriction was found to be more effective than exercise in reducing body weight. Exercise, on the other hand, was more effective in reducing adipose depots and muscle triglycerides, favorably altering muscle/liver desaturase activity and improving insulin sensitivity.

Segmentation and quantification of intra-ventricular/cerebral hemorrhage in CT scans by modified distance regularized level set evolution technique

PURPOSE

An automatic, accurate and fast segmentation of hemorrhage in brain Computed Tomography (CT) images is necessary for quantification and treatment planning when assessing a large number of data sets. Though manual segmentation is accurate, it is time consuming and tedious. Semi-automatic methods need user interactions and might introduce variability in results. Our study proposes a modified distance regularized level set evolution (MDRLSE) algorithm for hemorrhage segmentation.

METHODS

Study data set (from the ongoing CLEAR-IVH phase III clinical trial) is comprised of 200 sequential CT scans of 40 patients collected at 10 different hospitals using different machines/vendors. Data set contained both constant and variable slice thickness scans. Our study included pre-processing (filtering and skull removal), segmentation (MDRLSE which is a two-stage method with shrinking and expansion) with modified parameters for faster convergence and higher accuracy and post-processing (reduction in false positives and false negatives).

RESULTS

Results are validated against the gold standard marked manually by a trained CT reader and neurologist. Data sets are grouped as small, medium and large based on the volume of blood. Statistical analysis is performed for both training and test data sets in each group. The median Dice statistical indices (DSI) for the 3 groups are 0.8971, 0.8580 and 0.9173 respectively. Pre- and post-processing enhanced the DSI by 8 and 4% respectively.

CONCLUSIONS

The MDRLSE improved the accuracy and speed for segmentation and calculation of the hemorrhage volume compared to the original DRLSE method. The method generates quantitative information, which is useful for specific decision making and reduces the time needed for the clinicians to localize and segment the hemorrhagic regions.

A Simple and Fast Method of 3D Registration and Statistical Landmark Localization for Sparse Multi-Modal/Time-Series Neuroimages Based on Cortex Ellipse Fitting

Existing methods of neuroimage registration typically require high quality scans and are time-consuming. We propose a simple and fast method which allows intra-patient multi-modal and time-series neuroimage registration as well as landmark identification (including commissures and superior/inferior brain landmarks) for sparse data. The method is based on elliptical approximation of the brain cortical surface in the vicinity of the midsagittal plane (MSP). Scan registration is performed by a 3D affine transformation based on parameters of the cortex elliptical fit and by aligning the MSPs. The landmarks are computed using a statistical localization method based on analysis of 53 structural scans without detectable pathology. The method is illustrated for multi-modal registration, analysis of hemorrhagic stroke time series, and ischemic stroke follow ups, as well as for localization of hardly visible or not discernible landmarks in sparse neuroimages. The method also enables a statistical localization of landmarks in sparse morphological/non-morphological images, where landmark points may be invisible.

Extraction of the midsagittal plane from morphological neuroimages using the Kullback–Leibler’s measure

A theoretically simple and computationally efficient method to extract the midsagittal plane (MSP) from volumetric neuroimages is presented. The method works in two stages (coarse and fine) and is based on calculation of the Kullback and Leibler's (KL) measure, which characterizes the difference between two distributions. Slices along the sagittal direction are analyzed with respect to a reference slice to determine the coarse MSP. To calculate the final MSP, a local search algorithm is applied. The proposed method does not need any preprocessing, like reformatting, skull stripping, etc. The algorithm was validated quantitatively on 75 MRI datasets of different pulse sequences (T1WI, T2WI, FLAIR and SPGR) and MRA. The angular and distance errors between the calculated MSP and the ground truth lines marked by the expert were calculated. The average distance and angular deviation were 1.25 pixels and 0.63 degrees , respectively. In addition, the algorithm was tested qualitatively on PD, FLAIR, MRA, and CT datasets. To analyze the robustness of the method against rotation, inhomogeneity and noise, the phantom data were used.

Fast Talairach Transformation for magnetic resonance neuroimages

We introduce and validate the Fast Talairach Transformation (FTT). FTT is a rapid version of the Talairach transformation (TT) with the modified Talairach landmarks. Landmark identification is fully automatic and done in 3 steps: calculation of midsagittal plane, computing of anterior commissure (AC) and posterior commissure (PC) landmarks, and calculation of cortical landmarks. To perform these steps, we use fast and anatomy-based algorithms employing simple operations. FTT was validated for 215 diversified T1-weighted and spoiled gradient recalled (SPGR) MRI data sets. It calculates the landmarks and warps the Talairach-Tournoux atlas fully automatically in about 5 sec on a standard computer. The average distance errors in landmark localization are (in mm): 1.16 (AC), 1.49 (PC), 0.08 (left), 0.13 (right), 0.48 (anterior), 0.16 (posterior), 0.35 (superior), and 0.52 (inferior). Extensions to FTT by introducing additional landmarks and applying nonlinear warping against the ventricular system are addressed. Application of FTT to other brain atlases of anatomy, function, tracts, cerebrovasculature, and blood supply territories is discussed. FTT may be useful in a clinical setting and research environment: (1) when the TT is used traditionally, (2) when a global brain structure positioning with quick searching and labeling is required, (3) in urgent cases for quick image interpretation (eg, acute stroke), (4) when the difference between nonlinear and piecewise linear warping is negligible, (5) when automatic processing of a large number of cases is required, (6) as an initial atlas-scan alignment before performing nonlinear warping, and (7) as an initial atlas-guided segmentation of brain structures before further local processing.

Morphologic relationship among the corpus callosum, fornix, anterior commissure, and posterior commissure MRI-based variability study

RATIONALE AND OBJECTIVES

This study explores morphological relationships and structural variability of the corpus callosum (CC), fornix (Fo), anterior (AC), and posterior commissures (PC).

MATERIALS AND METHODS

These structures are extracted automatically on the midsagittal plane. The CC and Fo are modeled using best-fit ellipses. The parameters characterizing these structures and relationships among them are points, distances, angles, and eccentricities. The minimum, maximum and mean values, standard deviations, and coefficients of variation for all parameters are calculated for 62 diversified MRI datasets. Subsequently, the regression analysis and parameter distribution study are performed.

RESULTS

The parameters have at least 10% variations. The major axis of CC and eccentricities of CC and Fo vary much less than the other parameters The major axis of CC is approximately parallel to the AC-PC line. The mean eccentricity of each of CC and Fo is greater than 0.95. The most significant correlation (P < .05) is observed between various angles and the angle between the major axes of CC and Fo. The correlation is also significant between other angles and distances. The Weibull distribution characterizes the major axis of CC, and distance between the AC and the most superior point of CC. Distribution of angle between the major axes of CC and Fo is log (logistic), and normal for the AC-PC distance.

CONCLUSIONS

The AC-PC distance, used prevalently for brain normalization, is not correlated with any parameters except with the distance between the AC and the most superior point on the body of the CC with P < .05.

Rapid and automatic localization of the anterior and posterior commissure point landmarks in MR volumetric neuroimages

RATIONALE AND OBJECTIVE

Accurate identification of the anterior commissure (AC) and posterior commissure (PC) is critical in neuroradiology, functional neurosurgery, human brain mapping, and neuroscience research. Moreover, major stereotactic brain atlases are based on the AC and PC. Our goal is to provide an algorithm for a rapid, robust, accurate and automatic identification of AC and PC.

MATERIALS AND METHOD

The method exploits anatomical and radiological properties of AC, PC and surrounding structures, including morphological variability. The localization is done in two stages: coarse and fine. The coarse stage locates the AC and PC on the midsagittal plane by analyzing their relationships with the corpus callosum, fornix, and brainstem. The fine stage refines the AC and PC in a well-defined volume of interest, analyzing locations of lateral and third ventricles, interhemispheric fissure, and massa intermedia.

RESULTS

The algorithm was developed using simple operations, like histogramming, thresholding, region growing, 1D projections. It was tested on 94 diversified T1W and SPGR datasets. After the fine stage, 71 (76%) volumes had an error between 0-1 mm for the AC and 55 (59%) for the PC. The mean errors were 1.0 mm (AC) and 1.0 mm (PC). The accuracy has improved twice due to fine stage processing. The algorithm took about 1 second for coarse and 4 seconds for fine processing on P4, 2.5 GHz.

CONCLUSION

The use of anatomical and radiological knowledge including variability in algorithm formulation aids in localization of structures more accurately and robustly. This fully automatic algorithm is potentially useful in clinical setting and for research.

Dorsoventral Extension of the Talairach Transformation and Its Automatic Calculation for Magnetic Resonance Neuroimages

The Talairach transformation (TT), the most prevalent method for brain normalization and atlas-to-data warping, is conceptually simple, fast and can be automated. Two problems with the TT in the clinical setting that are addressed in this article are reduced accuracy at the orbitofrontal cortex and upper corpus callosum (CC) and unsuitability for functional neurosurgery because of incomplete scanning. To increase dorsoventral accuracy, we introduce 2 additional landmarks: the top of the CC (SM) and the most ventral point of the orbitofrontal cortex on the midsagittal slab (IM). A method for their automatic calculation is proposed and validated against 55 diversified magnetic resonance (MR) imaging cases. The SM and IM landmarks are identified accurately and robustly in an automatic way. The average error of SM localization is 0.69 mm, and 91% of all cases have an error not greater than 1 mm. The average error of IM localization is 0.98 mm, approximately three quarters of cases have an error not greater than 1 mm, and 95% of all cases have an error not larger than 2 mm. The SM is correlated (R(2) = 0.72) with the most superior cortical landmark, whereas the IM is only loosely correlated (R(2) = 0.22) with the most inferior cortical landmark. On average, the original TT overlays the atlas axial plate at -24 on the orbitofrontal cortex as opposed to the correct plate at -28. Therefore, 1-dimensional ventral scaling in the original TT is insufficient to cope with variability in the orbitofrontal cortex. The key advantages of our approach are the preserved conceptual simplicity of the TT, fully automatic identification of the new landmarks, improved accuracy of the atlas-to-data match without compromising performance, and enabled TT use in functional neurosurgery when a dorsal part of the brain is not available in the scan.

Fetal lung maturity analysis using ultrasound image features

This pilot study was carried out to find the feasibility of analyzing the maturity of the fetal lung using ultrasound images. Data were collected from normal pregnant women at intervals of two weeks from the gestation age of 24 to 38 weeks. Images were acquired at two centers located at different geographical locations. The total data acquired consisted of 750 images of immature and 250 images of mature class. A region of interest of 64 x 64 pixels was used for extracting the features. Various textural features were computed from the fetal lung and liver images. The ratios of fetal lung to liver feature values were investigated as possible indexes for classifying the images into those from mature (reduced pulmonary risk) and immature (possible pulmonary risk) lung. The features used are fractal dimension, lacunarity, and features derived from the histogram of the images. The following classifiers were used to classify the fetal lung images as belonging to mature or immature lung: nearest neighbor, k-nearest neighbor, modified k-nearest neighbor, multilayer perceptron, radial basis function network, and support vector machines. The classification accuracy obtained for the testing set ranges from 73% to 96%.