Weakly-supervised semantic segmentation (WSSS) methods, reliant on image-level labels indicating object presence, lack explicit correspondence between labels and regions of interest (ROIs), posing a significant challenge. Despite this, WSSS methods have attracted attention due to their much lower annotation costs compared to fully-supervised segmentation. Leveraging reinforcement learning (RL) self-play, we propose a novel WSSS method that gamifies image segmentation of a ROI. We formulate segmentation as a competition between two agents that compete to select ROI-containing patches until exhaustion of all such patches. The score at each time-step, used to compute the reward for agent training, represents likelihood of object presence within the selection, determined by an object presence detector pre-trained using only image-level binary classification labels of object presence. Additionally, we propose a game termination condition that can be called by either side upon exhaustion of all ROI-containing patches, followed by the selection of a final patch from each. Upon termination, the agent is incentivised if ROI-containing patches are exhausted or disincentivised if a ROI-containing patch is found by the competitor. This competitive setup ensures minimisation of over- or under-segmentation, a common problem with WSSS methods. Extensive experimentation across four datasets demonstrates significant performance improvements over recent state-of-the-art methods.

Obtaining feedback from time-constrained end-users is a major challenge in evaluating novel systems for specialised applications. The performance and feedback of engineers and surgeons was evaluated through an experiment where participants were asked to identify tumour locations within an anatomically realistic silicon liver model across three different conditions of an Augmented Reality (AR) prototype system (Baseline, Split AR and Full AR). Our findings show that engineers and surgeons share some similarities in their performance, feedback and behaviour, particularly when reliance on the AR system is high for both groups. However, engineers typically focus more on accuracy of the image alignment and are more accurate in their responses when supported by AR. Senior surgeons typically perform faster and use AR as supplementary information, while the performance of junior surgeons is more closely aligned to the performance of engineers. We conclude that engineers could be involved in preliminary evaluations of a surgical system or in evaluations of systems which are aimed at training junior surgeons, but that it is essential to involve surgeons in later evaluations, where ecological validity is a more important consideration.

Supervised machine learning-based medical image computing applications necessitate expert label curation, while unlabelled image data might be relatively abundant. Active learning methods aim to prioritise a subset of available image data for expert annotation, for label-efficient model training. We develop a controller neural network that measures priority of images in a sequence of batches, as in batch-mode active learning, for multi-class segmentation tasks. The controller is optimised by rewarding positive task-specific performance gain, within a Markov decision process (MDP) environment that also optimises the task predictor. In this work, the task predictor is a segmentation network. A meta-reinforcement learning algorithm is proposed with multiple MDPs, such that the pre-trained controller can be adapted to a new MDP that contains data from different institutes and/or requires segmentation of different organs or structures within the abdomen. We present experimental results using multiple CT datasets from more than one thousand patients, with segmentation tasks of nine different abdominal organs, to demonstrate the efficacy of the learnt prioritisation controller function and its cross-institute and cross-organ adaptability. We show that the proposed adaptable prioritisation metric yields converging segmentation accuracy for a new kidney segmentation task, unseen in training, using between approximately 40% to 60% of labels otherwise required with other heuristic or random prioritisation metrics. For clinical datasets of limited size, the proposed adaptable prioritisation offers a performance improvement of 22.6% and 10.2% in Dice score, for tasks of kidney and liver vessel segmentation, respectively, compared to random prioritisation and alternative active sampling strategies.

Minimally-invasive surgery (MIS) and robot-assisted minimally invasive (RAMIS) surgery offer well-documented benefits to patients such as reduced post-operative pain and shorter hospital stays.However, the automation of MIS and RAMIS through the use of AI has been slow due to difficulties in data acquisition and curation, partially caused by the ethical considerations of training, testing and deploying AI models in medical environments.We introduce \texttt{SARAMIS}, the first large-scale dataset of anatomically derived 3D rendering assets of the human abdominal anatomy.Using previously existing, open-source CT datasets of the human anatomy, we derive novel 3D meshes, tetrahedral volumes, textures and diffuse maps for over 104 different anatomical targets in the human body, representing the largest, open-source dataset of 3D rendering assets for synthetic simulation of vision tasks in MIS+RAMIS, increasing the availability of openly available 3D meshes in the literature by three orders of magnitude.We supplement our dataset with a series of GPU-enabled rendering environments, which can be used to generate datasets for realistic MIS/RAMIS tasks.Finally, we present an example of the use of \texttt{SARAMIS} assets for an autonomous navigation task in colonoscopy from CT abdomen-pelvis scans for the first time in the literature.\texttt{SARAMIS} is publically made available at https://github.com/NMontanaBrown/saramis/, with assets released under a CC-BY-NC-SA license.

Augmented Reality (AR) is considered to be a promising technology for the guidance of laparoscopic liver surgery. By overlaying pre-operative 3D information of the liver and internal blood vessels on the laparoscopic view, surgeons can better understand the location of critical structures. In an effort to enable AR, several authors have focused on the development of methods to obtain an accurate alignment between the laparoscopic video image and the pre-operative 3D data of the liver, without assessing the benefit that the resulting overlay can provide during surgery. In this paper, we present a study that aims to assess quantitatively and qualitatively the value of an AR overlay in laparoscopic surgery during a simulated surgical task on a phantom setup. We design a study where participants are asked to physically localise pre-operative tumours in a liver phantom using three image guidance conditions - a baseline condition without any image guidance, a condition where the 3D surfaces of the liver are aligned to the video and displayed on a black background, and a condition where video see-through AR is displayed on the laparoscopic video. Using data collected from a cohort of 24 participants which include 12 surgeons, we observe that compared to the baseline, AR decreases the median localisation error of surgeons on non-peripheral targets from 25.8 mm to 9.2 mm. Using subjective feedback, we also identify that AR introduces usability improvements in the surgical task and increases the perceived confidence of the users. Between the two tested displays, the majority of participants preferred to use the AR overlay instead of navigated view of the 3D surfaces on a separate screen. We conclude that AR has the potential to improve performance and decision making in laparoscopic surgery, and that improvements in overlay alignment accuracy and depth perception should be pursued in the future.

Ultrasound imaging is a commonly used technology for visualising patient anatomy in real-time during diagnostic and therapeutic procedures. High operator dependency and low reproducibility make ultrasound imaging and interpretation challenging with a steep learning curve. Automatic image classification using deep learning has the potential to overcome some of these challenges by supporting ultrasound training in novices, as well as aiding ultrasound image interpretation in patient with complex pathology for more experienced practitioners. However, the use of deep learning methods requires a large amount of data in order to provide accurate results. Labelling large ultrasound datasets is a challenging task because labels are retrospectively assigned to 2D images without the 3D spatial context available in vivo or that would be inferred while visually tracking structures between frames during the procedure. In this work, we propose a multi-modal convolutional neural network (CNN) architecture that labels endoscopic ultrasound (EUS) images from raw verbal comments provided by a clinician during the procedure. We use a CNN composed of two branches, one for voice data and another for image data, which are joined to predict image labels from the spoken names of anatomical landmarks. The network was trained using recorded verbal comments from expert operators. Our results show a prediction accuracy of 76% at image level on a dataset with 5 different labels. We conclude that the addition of spoken commentaries can increase the performance of ultrasound image classification, and eliminate the burden of manually labelling large EUS datasets necessary for deep learning applications.

Laparoscopic Ultrasound (LUS) is recommended as a standard-of-care when performing laparoscopic liver resections as it images sub-surface structures such as tumours and major vessels. Given that LUS probes are difficult to handle and some tumours are iso-echoic, registration of LUS images to a pre-operative CT has been proposed as an image-guidance method. This registration problem is particularly challenging due to the small field of view of LUS, and usually depends on both a manual initialisation and tracking to compose a volume, hindering clinical translation. In this paper, we extend a proposed registration approach using Content-Based Image Retrieval (CBIR), removing the requirement for tracking or manual initialisation. Pre-operatively, a set of possible LUS planes is simulated from CT and a descriptor generated for each image. Then, a Bayesian framework is employed to estimate the most likely sequence of CT simulations that matches a series of LUS images. We extend our CBIR formulation to use multiple labelled objects and constrain the registration by separating liver vessels into portal vein and hepatic vein branches. The value of this new labeled approach is demonstrated in retrospective data from 5 patients. Results show that, by including a series of 5 untracked images in time, a single LUS image can be registered with accuracies ranging from 5.7 to 16.4 mm with a success rate of 78%. Initialisation of the LUS to CT registration with the proposed framework could potentially enable the clinical translation of these image fusion techniques.

Automatic segmentation of abdominal anatomy on computed tomography (CT) images can support diagnosis, treatment planning, and treatment delivery workflows. Segmentation methods using statistical models and multi-atlas label fusion (MALF) require inter-subject image registrations, which are challenging for abdominal images, but alternative methods without registration have not yet achieved higher accuracy for most abdominal organs. We present a registration-free deep-learning-based segmentation algorithm for eight organs that are relevant for navigation in endoscopic pancreatic and biliary procedures, including the pancreas, the gastrointestinal tract (esophagus, stomach, and duodenum) and surrounding organs (liver, spleen, left kidney, and gallbladder). We directly compared the segmentation accuracy of the proposed method to the existing deep learning and MALF methods in a cross-validation on a multi-centre data set with 90 subjects. The proposed method yielded significantly higher Dice scores for all organs and lower mean absolute distances for most organs, including Dice scores of 0.78 versus 0.71, 0.74, and 0.74 for the pancreas, 0.90 versus 0.85, 0.87, and 0.83 for the stomach, and 0.76 versus 0.68, 0.69, and 0.66 for the esophagus. We conclude that the deep-learning-based segmentation represents a registration-free method for multi-organ abdominal CT segmentation whose accuracy can surpass current methods, potentially supporting image-guided navigation in gastrointestinal endoscopy procedures.