Globally, liver cancer causes more than 700,000 deaths each year and is the second-leading cause of death from cancer. Hepatocellular carcinoma (HCC) is the most common type of liver cancer in adults and accounts for most deaths in cirrhosis patients. Patients with early-stage liver cancer can be treated by surgical intervention with a good prognosis; thus, early diagnosis, as confirmed by liver pathology examination, is necessary to combat HCC. Conventional manual pathology examination requires considerable time and labor, even with established expertise. It is widely accepted that intelligent classifiers may prove effective in the diagnosis process. In this study, we used a GoogLeNet (Inception-V1)-based binary classifier to classify HCC histopathology images. The classifier achieved 91.37% (±2.49) accuracy, 92.16% (±4.93) sensitivity, and 90.57% (±2.54) specificity in HCC classification. Although the classification accuracy of deep learning is reported to be positively correlated with the amount of training data, it is often uncertain how much training data are required for deep learning to achieve satisfactory performance in clinical diagnosis. Moreover, deep learning methods require annotated data to generate efficient models. However, annotated data are a relatively scarce resource and can be expensive to obtain. Hence, the relationship between classification accuracy and the number of liver histopathology images for training was investigated. An inverse power law function-based estimation model is proposed to evaluate the minimum number of annotated training images required for a desired diagnostic accuracy.
This research devises a method to match marker-less surfaces for liver image registration. Surgeons usually glean preoperative liver information, such as anatomy and the locations of liver tumors or large intrahepatic vessels, from preoperative liver images that are obtained using Computed Tomography (CT) scans, Magnetic Resonance Imaging (MRI) or ultrasound. During minimally invasive surgery, surgeons use a laparoscope to obtain information about the intraoperative liver surface and identify an intraoperative liver tumor or the locations of vessels using the preoperative information. However, the liver can be lifted, shifted, flipped, squeezed or turned over during surgery. These manual operations can lead to severe deformation, so it is difficult to identify the location of intraoperative liver tumors or vessels. It is also difficult to accurately remove a liver tumor while avoiding injury to large intrahepatic vessels. This research proposes a method to determine the location of intraoperative vessels or tumors. The proposed method uses CT scans to construct a preoperative biomechanical volume model and uses a novel surface matching method to determine the relationship between the preoperative and intraoperative surfaces. The preoperative volume model is deformed by the finite element model in terms of the relationship that is defined by the proposed surface matching method, so that it aligns with the intraoperative surface model and shows the location of intraoperative vessels and tumors. The method of target registration error is measured for an ex vivo porcine liver to validate the proposed method. The results show that the error in the internal marker (which represents the location of the tumor and the vessel) is 4.54 ± 3.55 mm and the error in the surface marker is 2.98 ± 1.09 mm, which demonstrates the feasibility and high degree of accuracy of the proposed method.
In this paper, we present an integrated robotic arm with a flexible endoscope for laparoscopy. The endoscope holder is built to mimic a human operator that reacts to the surgeon's push while maintaining both the incision opening through the patient's body and the center of the endoscopic image. An impedance control algorithm is used to react to the surgeon's push when the robotic arm gets in the way. A modified software remote center-of-motion (RCM) constraint formulation then enables simultaneous RCM and impedance control. We derived the kinematic relationship between the robotic arm and line of sight of the flexible endoscope for image center control. Using this kinematic model, we integrated the task control for RCM and surgeon cooperation and the endoscope image centering into a semi-autonomous system. Implementation of the control algorithm with both matlab simulation and the HIWIN RA605-710 robotic arm with a MitCorp F500 flexible endoscope demonstrated the feasibility of the proposed algorithm.
7 January 2022 in Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization
The objective is to validate lung image registration using the featured surface matching method. The method can be applied to volume-to-surface deformable image registration, which is the 3D volume image to be deformed and aligned with a 3D surface image. Curvature variation extracts the surface feature. A group of neighbour curvature variations forms a local surface feature. A surface-matching scheme uses the local surface feature to determine the correspondence between two surface points. A finite element model uses the surface point correspondence to propagate the surface displacement into internal tissues and determine the intraoperative landmark locations from the deformed image. The validation uses the publicly available DIRLAB and POPI datasets to assess target registration errors. The work shows that the target registration error is 1.88 ± 1.16 mm, 4.77 ± 2.59 mm, and 4.12 ± 2.22 mm, and the initial error is 4.01 ± 2.91 mm, 11.10 ± 6.98 mm, 11.66 ± 6.23 mm for DIRLAB lung #1, DIRLAB lung #6 and POPI lung #1, respectively. The proposed method outperforms the average of other methods. The proposed novel surface matching method is feasible and validated.