Fairness and bias correction in machine learning for depression prediction across four study populations.

A significant level of stigma and inequality exists in mental healthcare, especially in under-served populations. Inequalities are reflected in the data collected for scientific purposes. When not properly accounted for, machine learning (ML) models learned from data can reinforce these structural inequalities or biases. Here, we present a systematic study of bias in ML models designed to predict depression in four different case studies covering different countries and populations. We find that standard ML approaches regularly present biased behaviors. We also show that mitigation techniques, both standard and our own post-hoc method, can be effective in reducing the level of unfair bias. There is no one best ML model for depression prediction that provides equality of outcomes. This emphasizes the importance of analyzing fairness during model selection and transparent reporting about the impact of debiasing interventions. Finally, we also identify positive habits and open challenges that practitioners could follow to enhance fairness in their models.

TopCoW: Benchmarking Topology-Aware Anatomical Segmentation of the Circle of Willis (CoW) for CTA and MRA.

The Circle of Willis (CoW) is an important network of arteries connecting major circulations of the brain. Its vascular architecture is believed to affect the risk, severity, and clinical outcome of serious neuro-vascular diseases. However, characterizing the highly variable CoW anatomy is still a manual and time-consuming expert task. The CoW is usually imaged by two angiographic imaging modalities, magnetic resonance angiography (MRA) and computed tomography angiography (CTA), but there exist limited public datasets with annotations on CoW anatomy, especially for CTA. Therefore we organized the TopCoW Challenge in 2023 with the release of an annotated CoW dataset. The TopCoW dataset was the first public dataset with voxel-level annotations for thirteen possible CoW vessel components, enabled by virtual-reality (VR) technology. It was also the first large dataset with paired MRA and CTA from the same patients. TopCoW challenge formalized the CoW characterization problem as a multiclass anatomical segmentation task with an emphasis on topological metrics. We invited submissions worldwide for the CoW segmentation task, which attracted over 140 registered participants from four continents. The top performing teams managed to segment many CoW components to Dice scores around 90%, but with lower scores for communicating arteries and rare variants. There were also topological mistakes for predictions with high Dice scores. Additional topological analysis revealed further areas for improvement in detecting certain CoW components and matching CoW variant topology accurately. TopCoW represented a first attempt at benchmarking the CoW anatomical segmentation task for MRA and CTA, both morphologically and topologically.

Deep Learning Segmentation of the Right Ventricle in Cardiac MRI: The M&Ms Challenge.

In recent years, several deep learning models have been proposed to accurately quantify and diagnose cardiac pathologies. These automated tools heavily rely on the accurate segmentation of cardiac structures in MRI images. However, segmentation of the right ventricle is challenging due to its highly complex shape and ill-defined borders. Hence, there is a need for new methods to handle such structure's geometrical and textural complexities, notably in the presence of pathologies such as Dilated Right Ventricle, Tricuspid Regurgitation, Arrhythmogenesis, Tetralogy of Fallot, and Inter-atrial Communication. The last MICCAI challenge on right ventricle segmentation was held in 2012 and included only 48 cases from a single clinical center. As part of the 12th Workshop on Statistical Atlases and Computational Models of the Heart (STACOM 2021), the M&Ms-2 challenge was organized to promote the interest of the research community around right ventricle segmentation in multi-disease, multi-view, and multi-center cardiac MRI. Three hundred sixty CMR cases, including short-axis and long-axis 4-chamber views, were collected from three Spanish hospitals using nine different scanners from three different vendors, and included a diverse set of right and left ventricle pathologies. The solutions provided by the participants show that nnU-Net achieved the best results overall. However, multi-view approaches were able to capture additional information, highlighting the need to integrate multiple cardiac diseases, views, scanners, and acquisition protocols to produce reliable automatic cardiac segmentation algorithms.

Telemedicine and Digital Health Applications in Vascular Surgery.

BACKGROUND: Telemedicine has the potential to revolutionize healthcare. While the development of digital health technologies for the management of patients with cardiovascular diseases has been increasingly reported, applications in vascular surgery have been far less specifically investigated. The aim of this review is to summarize applications related to telemedicine in vascular surgery, highlighting expected benefits, current limits and future directions. METHODS: The MEDLINE database was searched using a combination of keywords to identify studies related to telehealth/telemedicine in three main pathologies, including aortic, peripheral artery and carotid disease. A comprehensive literature review was performed to identify the type of digital application, intended use, expected benefits, strengths and limitations. RESULTS: Telemedicine can improve the management of patients through digital platforms allowing teleconsultation, telemonitoring or telecoaching. Intended use involved remote consultation with a vascular surgeon, applications to enhance education, self-management, follow-up or adherence to treatment or lifestyle changes. CONCLUSION: Telemedicine offers innovative perspectives to improve access to care in distant locations and optimize care through patients' empowerment and personalized follow-up, contributing to the development of precision medicine. Huge efforts remain necessary for its implementation in daily clinical practice and involve ethical, legal, technical, economic and cultural considerations.

Multi-task deep learning for glaucoma detection from color fundus images.

Glaucoma is an eye condition that leads to loss of vision and blindness if not diagnosed in time. Diagnosis requires human experts to estimate in a limited time subtle changes in the shape of the optic disc from retinal fundus images. Deep learning methods have been satisfactory in classifying and segmenting diseases in retinal fundus images, assisting in analyzing the increasing amount of images. Model training requires extensive annotations to achieve successful generalization, which can be highly problematic given the costly expert annotations. This work aims at designing and training a novel multi-task deep learning model that leverages the similarities of related eye-fundus tasks and measurements used in glaucoma diagnosis. The model simultaneously learns different segmentation and classification tasks, thus benefiting from their similarity. The evaluation of the method in a retinal fundus glaucoma challenge dataset, including 1200 retinal fundus images from different cameras and medical centers, obtained a [Formula: see text] AUC performance compared to an [Formula: see text] obtained by the same backbone network trained to detect glaucoma. Our approach outperforms other multi-task learning models, and its performance pairs with trained experts using [Formula: see text] times fewer parameters than training each task separately. The data and the code for reproducing our results are publicly available.

Vessel-CAPTCHA: An efficient learning framework for vessel annotation and segmentation.

Deep learning techniques for 3D brain vessel image segmentation have not been as successful as in the segmentation of other organs and tissues. This can be explained by two factors. First, deep learning techniques tend to show poor performances at the segmentation of relatively small objects compared to the size of the full image. Second, due to the complexity of vascular trees and the small size of vessels, it is challenging to obtain the amount of annotated training data typically needed by deep learning methods. To address these problems, we propose a novel annotation-efficient deep learning vessel segmentation framework. The framework avoids pixel-wise annotations, only requiring weak patch-level labels to discriminate between vessel and non-vessel 2D patches in the training set, in a setup similar to the CAPTCHAs used to differentiate humans from bots in web applications. The user-provided weak annotations are used for two tasks: (1) to synthesize pixel-wise pseudo-labels for vessels and background in each patch, which are used to train a segmentation network, and (2) to train a classifier network. The classifier network allows to generate additional weak patch labels, further reducing the annotation burden, and it acts as a second opinion for poor quality images. We use this framework for the segmentation of the cerebrovascular tree in Time-of-Flight angiography (TOF) and Susceptibility-Weighted Images (SWI). The results show that the framework achieves state-of-the-art accuracy, while reducing the annotation time by ∼77% w.r.t. learning-based segmentation methods using pixel-wise labels for training.

One-class autoencoder approach for optimal electrode set identification in wearable EEG event monitoring.

A limiting factor towards the wide use of wearable devices for continuous healthcare monitoring is their cumbersome and obtrusive nature. This is particularly true in electroencephalography (EEG), where numerous electrodes are placed in contact with the scalp to perform brain activity recordings. In this work, we propose to identify the optimal wearable EEG electrode set, in terms of minimal number of electrodes, comfortable location and performance, for EEG-based event detection and monitoring. By relying on the demonstrated power of autoencoder (AE) networks to learn latent representations from high-dimensional data, our proposed strategy trains an AE architecture in a one-class classification setup with different electrode combinations as input data. The model performance is assessed using the F-score. Alpha waves detection is the use case through which we demonstrate that the proposed method allows to detect a brain state from an optimal set of electrodes. The so-called wearable configuration, consisting of electrodes in the forehead and behind the ear, is the chosen optimal set, with an average F-score of 0.78. This study highlights the beneficial impact of a learning-based approach in the design of wearable devices for real-life event-related monitoring.

Substantially thinner internal granular layer and reduced molecular layer surface in the cerebellar cortex of the Tc1 mouse model of down syndrome - a comprehensive morphometric analysis with active staining contrast-enhanced MRI.

Down Syndrome is a chromosomal disorder that affects the development of cerebellar cortical lobules. Impaired neurogenesis in the cerebellum varies among different types of neuronal cells and neuronal layers. In this study, we developed an imaging analysis framework that utilizes gadolinium-enhanced ex vivo mouse brain MRI. We extracted the middle Purkinje layer of the mouse cerebellar cortex, enabling the estimation of the volume, thickness, and surface area of the entire cerebellar cortex, the internal granular layer, and the molecular layer in the Tc1 mouse model of Down Syndrome. The morphometric analysis of our method revealed that a larger proportion of the cerebellar thinning in this model of Down Syndrome resided in the inner granule cell layer, while a larger proportion of the surface area shrinkage was in the molecular layer.

Autism spectrum disorder characterization in children by capturing local-regional brain changes in MRI.

PURPOSE: To design a multiscale descriptor capable of capturing complex local-regional unfolding patterns to support quantitation and diagnosis of autism spectrum disorders (ASD) using T1-weighted structural magnetic resonance images (MRI) with voxel size of 1 × 1 × 1 mm. METHODS: The proposed image descriptor uses an adapted multiscale representation, the Curvelet transform, interpretable in terms of texture (local) and shape (regional) to characterize brain regions, and a Generalized Gaussian Distribution (GGD) to reduce feature dimensionality. In this approach, each MRI is first parcelled into 3D anatomical regions. Each resultant region is represented by a single 2D image where slices are placed next to each other. Each 2D image is characterized by mapping it to the Curvelet space and each of the different Curvelet sub-bands is described by the set of GGD parameters. To assess the discriminant power of the proposed descriptor, a classification model per brain region was built to differentiate ASD patients from control subjects. Models were constructed with support vector machines and evaluated using two samples from heterogeneous databases, namely Autism Brain Imaging Data Exchange - ABIDE I (34 ASD and 34 controls, mean age 11.46 ± 2.03 and 11.53 ± 1.79 yr, respectively, male population) and ABIDE II (42 ASD and 41 controls, mean age 10.09 ± 1.37 and 10.52 ± 1.27 yr, respectively, male population), for a total of 151 individuals. RESULTS: When the model was trained with ABIDE II sample and tested with ABIDE I on a hold-out validation, an area under receiver operator curve (AUC) of 0.69 was computed. When each sample was independently used under a cross-validation scheme, the estimated AUC was 0.75 ± 0.02 for ABIDE I and 0.77 ± 0.01 for ABIDE II. This analysis determined a set of discriminant regions widely reported in the literature as characteristic of ASD. CONCLUSIONS: The presented image descriptor demonstrated differences at local and regional level when high differences were observed in the Curvelet sub-bands. The method is simple in conceptual terms, robust to several sources of noise, and has a very low computational cost.

DeepIGeoS: A Deep Interactive Geodesic Framework for Medical Image Segmentation.

Accurate medical image segmentation is essential for diagnosis, surgical planning and many other applications. Convolutional Neural Networks (CNNs) have become the state-of-the-art automatic segmentation methods. However, fully automatic results may still need to be refined to become accurate and robust enough for clinical use. We propose a deep learning-based interactive segmentation method to improve the results obtained by an automatic CNN and to reduce user interactions during refinement for higher accuracy. We use one CNN to obtain an initial automatic segmentation, on which user interactions are added to indicate mis-segmentations. Another CNN takes as input the user interactions with the initial segmentation and gives a refined result. We propose to combine user interactions with CNNs through geodesic distance transforms, and propose a resolution-preserving network that gives a better dense prediction. In addition, we integrate user interactions as hard constraints into a back-propagatable Conditional Random Field. We validated the proposed framework in the context of 2D placenta segmentation from fetal MRI and 3D brain tumor segmentation from FLAIR images. Experimental results show our method achieves a large improvement from automatic CNNs, and obtains comparable and even higher accuracy with fewer user interventions and less time compared with traditional interactive methods.

Inference of Cerebrovascular Topology With Geodesic Minimum Spanning Trees.

A vectorial representation of the vascular network that embodies quantitative features-location, direction, scale, and bifurcations-has many potential cardio- and neuro-vascular applications. We present VTrails, an end-to-end approach to extract geodesic vascular minimum spanning trees from angiographic data by solving a connectivity-optimized anisotropic level-set over a voxel-wise tensor field representing the orientation of the underlying vasculature. Evaluating real and synthetic vascular images, we compare VTrails against the state-of-the-art ridge detectors for tubular structures by assessing the connectedness of the vesselness map and inspecting the synthesized tensor field. The inferred geodesic trees are then quantitatively evaluated within a topologically aware framework, by comparing the proposed method against popular vascular segmentation tool kits on clinical angiographies. VTrails potentials are discussed towards integrating groupwise vascular image analyses. The performance of VTrails demonstrates its versatility and usefulness also for patient-specific applications in interventional neuroradiology and vascular surgery.

Interactive Medical Image Segmentation Using Deep Learning With Image-Specific Fine Tuning.

Convolutional neural networks (CNNs) have achieved state-of-the-art performance for automatic medical image segmentation. However, they have not demonstrated sufficiently accurate and robust results for clinical use. In addition, they are limited by the lack of image-specific adaptation and the lack of generalizability to previously unseen object classes (a.k.a. zero-shot learning). To address these problems, we propose a novel deep learning-based interactive segmentation framework by incorporating CNNs into a bounding box and scribble-based segmentation pipeline. We propose image-specific fine tuning to make a CNN model adaptive to a specific test image, which can be either unsupervised (without additional user interactions) or supervised (with additional scribbles). We also propose a weighted loss function considering network and interaction-based uncertainty for the fine tuning. We applied this framework to two applications: 2-D segmentation of multiple organs from fetal magnetic resonance (MR) slices, where only two types of these organs were annotated for training and 3-D segmentation of brain tumor core (excluding edema) and whole brain tumor (including edema) from different MR sequences, where only the tumor core in one MR sequence was annotated for training. Experimental results show that: 1) our model is more robust to segment previously unseen objects than state-of-the-art CNNs; 2) image-specific fine tuning with the proposed weighted loss function significantly improves segmentation accuracy; and 3) our method leads to accurate results with fewer user interactions and less user time than traditional interactive segmentation methods.

Imaging the human placental microcirculation with micro-focus computed tomography: Optimisation of tissue preparation and image acquisition.

Micro-CT provides 3D volume imaging with spatial resolution at the micrometre scale. We investigated the optimal human placenta tissue preparation (contrast agent, perfusion pressure, perfusion location and perfusion vessel) and imaging (energy, target material, exposure time and frames) parameters. Microfil (Flow Tech, Carver, MA) produced better fill than Barium sulphate (84.1%(±11.5%)vs70.4%(±18.02%) p = 0.01). Perfusion via umbilical artery produced better fill than via chorionic vessels (83.8%(±17.7%)vs78.0%(±21.9%), p < 0.05), or via umbilical vein (83.8%(±16.4%)vs69.8%(±20.3%), p < 0.01). Imaging at 50 keV with a molybdenum target produced the best contrast to noise ratio. We propose this method to enable quantification and comparison of the human fetoplacental vascular tree.

Standardized unfold mapping: a technique to permit left atrial regional data display and analysis.

PURPOSE: Left atrial arrhythmia substrate assessment can involve multiple imaging and electrical modalities, but visual analysis of data on 3D surfaces is time-consuming and suffers from limited reproducibility. Unfold maps (e.g., the left ventricular bull's eye plot) allow 2D visualization, facilitate multimodal data representation, and provide a common reference space for inter-subject comparison. The aim of this work is to develop a method for automatic representation of multimodal information on a left atrial standardized unfold map (LA-SUM). METHODS: The LA-SUM technique was developed and validated using 18 electroanatomic mapping (EAM) LA geometries before being applied to ten cardiac magnetic resonance/EAM paired geometries. The LA-SUM was defined as an unfold template of an average LA mesh, and registration of clinical data to this mesh facilitated creation of new LA-SUMs by surface parameterization. RESULTS: The LA-SUM represents 24 LA regions on a flattened surface. Intra-observer variability of LA-SUMs for both EAM and CMR datasets was minimal; root-mean square difference of 0.008 ± 0.010 and 0.007 ± 0.005 ms (local activation time maps), 0.068 ± 0.063 gs (force-time integral maps), and 0.031 ± 0.026 (CMR LGE signal intensity maps). Following validation, LA-SUMs were used for automatic quantification of post-ablation scar formation using CMR imaging, demonstrating a weak but significant relationship between ablation force-time integral and scar coverage (R 2 = 0.18, P < 0.0001). CONCLUSIONS: The proposed LA-SUM displays an integrated unfold map for multimodal information. The method is applicable to any LA surface, including those derived from imaging and EAM systems. The LA-SUM would facilitate standardization of future research studies involving segmental analysis of the LA.

Investigating Cardiac Motion Patterns Using Synthetic High-Resolution 3D Cardiovascular Magnetic Resonance Images and Statistical Shape Analysis.

Diagnosis of ventricular dysfunction in congenital heart disease is more and more based on medical imaging, which allows investigation of abnormal cardiac morphology and correlated abnormal function. Although analysis of 2D images represents the clinical standard, novel tools performing automatic processing of 3D images are becoming available, providing more detailed and comprehensive information than simple 2D morphometry. Among these, statistical shape analysis (SSA) allows a consistent and quantitative description of a population of complex shapes, as a way to detect novel biomarkers, ultimately improving diagnosis and pathology understanding. The aim of this study is to describe the implementation of a SSA method for the investigation of 3D left ventricular shape and motion patterns and to test it on a small sample of 4 congenital repaired aortic stenosis patients and 4 age-matched healthy volunteers to demonstrate its potential. The advantage of this method is the capability of analyzing subject-specific motion patterns separately from the individual morphology, visually and quantitatively, as a way to identify functional abnormalities related to both dynamics and shape. Specifically, we combined 3D, high-resolution whole heart data with 2D, temporal information provided by cine cardiovascular magnetic resonance images, and we used an SSA approach to analyze 3D motion per se. Preliminary results of this pilot study showed that using this method, some differences in end-diastolic and end-systolic ventricular shapes could be captured, but it was not possible to clearly separate the two cohorts based on shape information alone. However, further analyses on ventricular motion allowed to qualitatively identify differences between the two populations. Moreover, by describing shape and motion with a small number of principal components, this method offers a fully automated process to obtain visually intuitive and numerical information on cardiac shape and motion, which could be, once validated on a larger sample size, easily integrated into the clinical workflow. To conclude, in this preliminary work, we have implemented state-of-the-art automatic segmentation and SSA methods, and we have shown how they could improve our understanding of ventricular kinetics by visually and potentially quantitatively highlighting aspects that are usually not picked up by traditional approaches.

Selection bias in the reported performances of AD classification pipelines.

The last decade has seen a great proliferation of supervised learning pipelines for individual diagnosis and prognosis in Alzheimer's disease. As more pipelines are developed and evaluated in the search for greater performance, only those results that are relatively impressive will be selected for publication. We present an empirical study to evaluate the potential for optimistic bias in classification performance results as a result of this selection. This is achieved using a novel, resampling-based experiment design that effectively simulates the optimisation of pipeline specifications by individuals or collectives of researchers using cross validation with limited data. Our findings indicate that bias can plausibly account for an appreciable fraction (often greater than half) of the apparent performance improvement associated with the pipeline optimisation, particularly in small samples. We discuss the consistency of our findings with patterns observed in the literature and consider strategies for bias reduction and mitigation.

Detecting Clinically Meaningful Shape Clusters in Medical Image Data: Metrics Analysis for Hierarchical Clustering Applied to Healthy and Pathological Aortic Arches.

OBJECTIVE: Today's growing medical image databases call for novel processing tools to structure the bulk of data and extract clinically relevant information. Unsupervised hierarchical clustering may reveal clusters within anatomical shape data of patient populations as required for modern precision medicine strategies. Few studies have applied hierarchical clustering techniques to three-dimensional patient shape data and results depend heavily on the chosen clustering distance metrics and linkage functions. In this study, we sought to assess clustering classification performance of various distance/linkage combinations and of different types of input data to obtain clinically meaningful shape clusters. METHODS: We present a processing pipeline combining automatic segmentation, statistical shape modeling, and agglomerative hierarchical clustering to automatically subdivide a set of 60 aortic arch anatomical models into healthy controls, two groups affected by congenital heart disease, and their respective subgroups as defined by clinical diagnosis. Results were compared with traditional morphometrics and principal component analysis of shape features. RESULTS: Our pipeline achieved automatic division of input shape data according to primary clinical diagnosis with high F-score (0.902 ± 0.042) and Matthews correlation coefficient (0.851 ± 0.064) using the correlation/weighted distance/linkage combination. Meaningful subgroups within the three patient groups were obtained and benchmark scores for automatic segmentation and classification performance are reported. CONCLUSION: Clustering results vary depending on the distance/linkage combination used to divide the data. Yet, clinically relevant shape clusters and subgroups could be found with high specificity and low misclassification rates. SIGNIFICANCE: Detecting disease-specific clusters within medical image data could improve image-based risk assessment, treatment planning, and medical device development in complex disease.

Automated multiple trajectory planning algorithm for the placement of stereo-electroencephalography (SEEG) electrodes in epilepsy treatment.

PURPOSE: About one-third of individuals with focal epilepsy continue to have seizures despite optimal medical management. These patients are potentially curable with neurosurgery if the epileptogenic zone (EZ) can be identified and resected. Stereo-electroencephalography (SEEG) to record epileptic activity with intracranial depth electrodes may be required to identify the EZ. Each SEEG electrode trajectory, the path between the entry on the skull and the cerebral target, must be planned carefully to avoid trauma to blood vessels and conflicts between electrodes. In current clinical practice trajectories are determined manually, typically taking 2-3 h per patient (15 min per electrode). Manual planning (MP) aims to achieve an implantation plan with good coverage of the putative EZ, an optimal spatial resolution, and 3D distribution of electrodes. Computer-assisted planning tools can reduce planning time by quantifying trajectory suitability. METHODS: We present an automated multiple trajectory planning (MTP) algorithm to compute implantation plans. MTP uses dynamic programming to determine a set of plans. From this set a depth-first search algorithm finds a suitable plan. We compared our MTP algorithm to (a) MP and (b) an automated single trajectory planning (STP) algorithm on 18 patient plans containing 165 electrodes. RESULTS: MTP changed all 165 trajectories compared to MP. Changes resulted in lower risk (122), increased grey matter sampling (99), shorter length (92), and surgically preferred entry angles (113). MTP changed 42 % (69/165) trajectories compared to STP. Every plan had between 1 to 8 (median 3.5) trajectories changed to resolve electrode conflicts, resulting in surgically preferred plans. CONCLUSION: MTP is computationally efficient, determining implantation plans containing 7-12 electrodes within 1 min, compared to 2-3 h for MP.

Automatic segmentation of right ventricle in cardiac cine MR images using a saliency analysis.

PURPOSE: Accurate measurement of the right ventricle (RV) volume is important for the assessment of the ventricular function and a biomarker of the progression of any cardiovascular disease. However, the high RV variability makes difficult a proper delineation of the myocardium wall. This paper introduces a new automatic method for segmenting the RV volume from short axis cardiac magnetic resonance (MR) images by a salient analysis of temporal and spatial observations. METHODS: The RV volume estimation starts by localizing the heart as the region with the most coherent motion during the cardiac cycle. Afterward, the ventricular chambers are identified at the basal level using the isodata algorithm, the right ventricle extracted, and its centroid computed. A series of radial intensity profiles, traced from this centroid, is used to search a salient intensity pattern that models the inner-outer myocardium boundary. This process is iteratively applied toward the apex, using the segmentation of the previous slice as a regularizer. The consecutive 2D segmentations are added together to obtain the final RV endocardium volume that serves to estimate also the epicardium. RESULTS: Experiments performed with a public dataset, provided by the RV segmentation challenge in cardiac MRI, demonstrated that this method is highly competitive with respect to the state of the art, obtaining a Dice score of 0.87, and a Hausdorff distance of 7.26 mm while a whole volume was segmented in about 3 s. CONCLUSIONS: The proposed method provides an useful delineation of the RV shape using only the spatial and temporal information of the cine MR images. This methodology may be used by the expert to achieve cardiac indicators of the right ventricle function.

Slic-Seg: A minimally interactive segmentation of the placenta from sparse and motion-corrupted fetal MRI in multiple views.

Segmentation of the placenta from fetal MRI is challenging due to sparse acquisition, inter-slice motion, and the widely varying position and shape of the placenta between pregnant women. We propose a minimally interactive framework that combines multiple volumes acquired in different views to obtain accurate segmentation of the placenta. In the first phase, a minimally interactive slice-by-slice propagation method called Slic-Seg is used to obtain an initial segmentation from a single motion-corrupted sparse volume image. It combines high-level features, online Random Forests and Conditional Random Fields, and only needs user interactions in a single slice. In the second phase, to take advantage of the complementary resolution in multiple volumes acquired in different views, we further propose a probability-based 4D Graph Cuts method to refine the initial segmentations using inter-slice and inter-image consistency. We used our minimally interactive framework to examine the placentas of 16 mid-gestation patients from MRI acquired in axial and sagittal views respectively. The results show the proposed method has 1) a good performance even in cases where sparse scribbles provided by the user lead to poor results with the competitive propagation approaches; 2) a good interactivity with low intra- and inter-operator variability; 3) higher accuracy than state-of-the-art interactive segmentation methods; and 4) an improved accuracy due to the co-segmentation based refinement, which outperforms single volume or intensity-based Graph Cuts.