Reconstructing the 3D geometry of the surgical site and detecting instruments within it are important tasks for surgical navigation systems and robotic surgery automation. Traditional approaches treat each problem in isolation and do not account for the intrinsic relationship between segmentation and stereo matching. In this paper, we present a learning-based framework that jointly estimates disparity and binary tool segmentation masks. The core component of our architecture is a shared feature encoder which allows strong interaction between the aforementioned tasks. Experimentally, we train two variants of our network with different capacities and explore different training schemes including both multi-task and single-task learning. Our results show that supervising the segmentation task improves our network’s disparity estimation accuracy. We demonstrate a domain adaptation scheme where we supervise the segmentation task with monocular data and achieve domain adaptation of the adjacent disparity task, reducing disparity End-Point-Error and depth mean absolute error by 77.73% and 61.73% respectively compared to the pre-trained baseline model. Our best overall multi-task model, trained with both disparity and segmentation data in subsequent phases, achieves 89.15% mean Intersection-over-Union in RIS and 3.18 millimetre depth mean absolute error in SCARED test sets. Our proposed multi-task architecture is real-time, able to process ( $1280\times 1024$ ) stereo input and simultaneously estimate disparity maps and segmentation masks at 22 frames per second. The model code and pre-trained models are made available: https://github.com/dimitrisPs/msdesis

Video streams are utilised to guide minimally-invasive surgery and diagnosis in a wide range of procedures, and many computer-assisted techniques have been developed to automatically analyse them. These approaches can provide additional information to the surgeon such as lesion detection, instrument navigation, or anatomy 3D shape modelling. However, the necessary image features to recognise these patterns are not always reliably detected due to the presence of irregular light patterns such as specular highlight reflections. In this paper, we aim at removing specular highlights from endoscopic videos using machine learning. We propose using a temporal generative adversarial network (GAN) to inpaint the hidden anatomy under specularities, inferring its appearance spatially and from neighbouring frames, where they are not present in the same location. This is achieved using in-vivo data from gastric endoscopy (Hyper Kvasir) in a fully unsupervised manner that relies on the automatic detection of specular highlights. System evaluations show significant improvements to other methods through direct comparison and ablation studies that depict the importance of the network's temporal and transfer learning components. The generalisability of our system to different surgical setups and procedures was also evaluated qualitatively on in-vivo data of gastric endoscopy and ex-vivo porcine data (SERV-CT, SCARED). We also assess the effect of our method in comparison to other methods on computer vision tasks that underpin 3D reconstruction and camera motion estimation, namely stereo disparity, optical flow, and sparse point feature matching. These are evaluated quantitatively and qualitatively and results show a positive effect of our specular inpainting method on these tasks in a novel comprehensive analysis. Our code and dataset are made available at https://github.com/endomapper/Endo-STTN.

Surgical instrument segmentation can be used in a range of computer assisted interventions and automation in surgical robotics. While deep learning architectures have rapidly advanced the robustness and performance of segmentation models, most are still reliant on supervision and large quantities of labelled data. In this paper, we present a novel method for surgical image generation that can fuse robotic instrument simulation and recent domain adaptation techniques to synthesize artificial surgical images to train surgical instrument segmentation models. We integrate attention modules into well established image generation pipelines and propose a novel cost function to support supervision from simulation frames in model training. We provide an extensive evaluation of our method in terms of segmentation performance along with a validation study on image quality using evaluation metrics. Additionally, we release a novel segmentation dataset from real surgeries that will be shared for research purposes. Both binary and semantic segmentation have been considered, and we show the capability of our synthetic images to train segmentation models compared with the latest methods from the literature.

We propose a new method to add an uncalibrated node into a network of calibrated cameras using only pairwise point correspondences. While previous methods perform this task using triple correspondences, these are often difficult to establish when there is limited overlap between different views. In such challenging cases we must rely on pairwise correspondences and our solution becomes more advantageous. Our method includes an 11-point minimal solution for the intrinsic and extrinsic calibration of a camera from pairwise correspondences with other two calibrated cameras, and a new inlier selection framework that extends the traditional RANSAC family of algorithms to sampling across multiple datasets. Our method is validated on different application scenarios where a lack of triple correspondences might occur: addition of a new node to a camera network; calibration and motion estimation of a moving camera inside a camera network; and addition of views with limited overlap to a Structure-from-Motion model.

Purpose: Fetoscopic laser photocoagulation is a minimally invasive surgical procedure used to treat twin-to-twin transfusion syndrome (TTTS), which involves localization and ablation of abnormal vascular connections on the placenta to regulate the blood flow in both fetuses. This procedure is particularly challenging due to the limited field of view, poor visibility, occasional bleeding, and poor image quality. Fetoscopic mosaicking can help in creating an image with the expanded field of view which could facilitate the clinicians during the TTTS procedure. Methods: We propose a deep learning-based mosaicking framework for diverse fetoscopic videos captured from different settings such as simulation, phantoms, ex vivo, and in vivo environments. The proposed mosaicking framework extends an existing deep image homography model to handle video data by introducing the controlled data generation and consistent homography estimation modules. Training is performed on a small subset of fetoscopic images which are independent of the testing videos. Results: We perform both quantitative and qualitative evaluations on 5 diverse fetoscopic videos (2400 frames) that captured different environments. To demonstrate the robustness of the proposed framework, a comparison is performed with the existing feature-based and deep image homography methods. Conclusion: The proposed mosaicking framework outperformed existing methods and generated meaningful mosaic, while reducing the accumulated drift, even in the presence of visual challenges such as specular highlights, reflection, texture paucity, and low video resolution.

We introduce a novel approach for learning a complex multi-stage needle pick-and-place manipulation task for surgical applications using Reinforcement Learning without expert demonstrations or explicit curriculum. The proposed method is based on a recursive decomposition of the original task into a sequence of sub-tasks with increasing complexity and utilizes an actor-critic algorithm with deterministic policy output. In this work, exploratory bottlenecks have been used by a human expert as convenient boundary points for partitioning complex tasks into simpler subunits. Our method has successfully learnt a policy for the needle pick-and-place task, whereas the state-of-the-art TD3+HER method is unable to achieve success without the help of expert demonstrations. Comparison results show that our method achieves the highest performance with a 91% average success rate.