Application of Computer-Assisted Surgery System Based on Artificial Intelligence in Pediatric Precise Oncological Surgery

Abstract

Pediatric oncological surgery is difficult and challenging, especially in children with malignant solid tumors. Compared with adults, children have immature organs, thin blood vessels, and poor surgical tolerance. Moreover, pediatric malignant solid tumors are often huge, complex in location, fast-growing, and highly malignant. With artificial intelligence and machine learning breaking through many bottlenecks, computer-assisted precision medicine has also taken a quantum leap forward. Ten years ago, Professor Dong’s group invented the Computer-assisted Surgery System (Hisense CAS). Now, this three-dimensional (3D) visualization technology based on artificial intelligence has been used for pediatric precise oncological surgery and has been upgraded to version 5.0. Hisense CAS was developed based on enhanced pediatric CT data, so it has advantages in displaying pediatric solid tumors. CAS can display the adjacent relationships of the tumor with the surrounding tissues (especially the compressed blood vessels) in a 3D, dynamic, and complete manner through rapid and accurate 3D reconstruction of organs, tumors, and blood vessels. Then, precise preoperative evaluations and surgical planning can be carried out. This chapter focuses on individualized computer-assisted surgical planning and progress in common and complex pediatric tumors (such as malignant liver tumors, retroperitoneal tumors, and mediastinal tumors) and introduces experience in improving the resectability of tumors and reducing surgical complications.

1. Introduction

With decreasing mortality from infectious diseases and increasing cure rates for congenital malformations, pediatric tumors have become an important disease factor endangering the health of children, and their incidence continues to rise [1, 2]. Surgical resection is the most effective and important treatment for the eradication of pediatric solid tumors, especially malignant solid tumors. However, pediatric tumors are often diverse, complex in location, and large in size compared to tumors in the young body. Malignant tumors are highly malignant and grow rapidly. Moreover, compared with that of adults, the organ structure of pediatric patients is slender and has poor tolerance, thus the space for surgical treatment is limited. Therefore, there is a greater need for high-technology tools that help perform precise and meticulous surgical procedures [3].

Technological innovation and interdisciplinary integration have brought surgery to a brand new stage, namely, the era of precision surgery. Precision surgery is a whole process of surgery-centered surgical practices, covering all stages from disease assessment, clinical decision-making, surgical planning, and surgical resection to perioperative management. Computer-assisted surgery (CAS), a typical representative of medical-industrial integration, is a new technology based on artificial intelligence and machine learning that can process and learn large amounts of medical data and information at high speed and then provide technical support to surgeons through a virtual surgical environment to assist in the realization of precision surgery [4].

2. Methodology

This chapter provides a retrospective summary of the key technologies of our self-developed computer-assisted surgery system “Hisense CAS” and analyses its practical application in pediatric oncological surgery, providing individualized computer-assisted surgical planning for common and complex pediatric tumors (such as malignant tumors of the liver, retroperitoneal tumors, and mediastinal tumors) to improve tumor resectability and reduce surgery-related complications [5, 6].

3. Computer-assisted surgery

CAS is a new surgical concept that refers to the use of computer technology for presurgical planning and to guide or assist surgical procedures. CAS is generally considered to include (i) creation of virtual images of patients; (ii) analysis and in-depth processing of patient images; (iii) diagnosis, presurgical planning, and simulation of surgical steps; (iv) surgical navigation; and (v) robotic surgery. With its development and use in the medical field, CAS has helped realize precision surgery.

3.1 Digital three-dimensional reconstruction and simulation surgery

The technical basis of digital three-dimensional (3D) reconstruction is to convert two-dimensional (2D) cross-sectional images such as CT or MRI into 3D visual images using computer algorithms to provide the operator with more intuitive stereoscopic images for diagnosis and preoperative evaluation. Through the virtual reality surgeries available through modern computer technology, a virtual surgery model for specific individualized surgical modality evaluations can be established. The surgeon can input the conceived surgical plan into the computer, combine it with the presurgical medical images, and form a three-dimensional image after processing by the software system to understand in detail the specific location, involvement range, and adjacent relationships of the tumor, especially the involvement of blood vessels. Medical image data and virtual surgery systems are also used to reasonably customize individualized surgical plans to reduce surgical injury, avoid damage to surrounding tissues, improve the precision of lesion localization, and increase the success rate of surgery [7].

3.2 Research and development of CAS based on artificial intelligence

The Hisense Computer-assisted Surgery System (Hisense CAS) was developed by Prof. Dong’s group in 2013. This system can perform medical image preprocessing with CT imaging DICOM data, especially low-quality data that can be enhanced in high definition before preprocessing. The medical images labeled with features such as greyscale and texture features are then subjected to deep machine learning by U-Net on a large number of standard DICOM files. Thus, automatic and accurate segmentation of new input data is achieved. The segmentation results are processed by filtering, CT interlayer adaptive correspondence point interpolation, morphology, pattern recognition, and other algorithms, and a self-learning topological model is established to model and track the vessel shape in the three phases (arterial phase, venous phase, and balance phase) of imaging. Then, the matching cubes and ray cast algorithms are used for color rendering, and finally, the 3D alignment algorithm is used to stereoscopically align the three-phase data to accurately obtain enhanced 3D images visualizing the target organs, lesions, and blood vessels. This system can precisely observe the relationships of the lesion with blood vessels and organs in 3D, calculate the volume of the organs, lesions, and blood supply area of each blood vessel branch, perform virtual surgical resection, and determine the best surgical resection line.

With the progression of clinical needs, artificial intelligence technology, and machine learning based on big data, Hisense CAS has been updated to version 5.0. The improved algorithm enables less manual operation and a 25–30% reduction in 3D reconstruction time, and the whole process takes approximately 20 minutes. Hisense CAS can reconstruct more than 4 levels of vessels and distinguish tumors and vessels with 0.5 cm spacing. The Dice value of solid tissues can reach more than 95%, and that of ducts can reach more than 90%. Hisense CAS can also display the overall 3D anatomical relationship and pipeline variations in a semitransparent and interactive way, calculate the distance between any two points and the angle of travel of any blood vessel, the range of innervation or drainage, and the volume of organs and tumors, and provide other information that cannot be obtained from traditional 2D images. In addition, a cloud-based 3D visualization platform for precision surgery based on B/S architecture was constructed to realize the data interactions between the PC terminal browser and CAS and to store, manage and share 2D and 3D image data (Figure 1) [8, 9].

Figure 1.

Hisense computer-assisted surgery system based on artificial intelligence.

3.3 Gesture control intelligent display module (Hisense SID)

Real-time surgical navigation is used to accurately correspond the preoperative image with intraoperative organ anatomy, and through instruments or signal transmission, real-time feedback to the image is provided to reconstruct the model, enable clear positioning, and achieve precision surgery. The 3D image gesture control intelligent display module (Hisense SID) developed by Prof. Dong’s group, based on somatosensory interaction, motion capture, and other technologies, can quickly and precisely realize human-computer interactions through simple gesture operations within a specific range to ensure intraoperative sterility. The operator can rotate, zoom in and zoom out on the digital 3D image through different gesture control commands to view the required details. In addition, the system provides intelligent tracking of specific operators, thus eliminating interference from surrounding personnel, reducing misuse, and improving recognition rates (Figure 2) [10].

Figure 2.

Real-time surgical navigation through Hisense SID gesture intelligent control.

4. Application of CAS in pediatric precise oncological surgery

According to a survey by the International Society of Pediatric Oncology, the incidence of pediatric tumors has increased at a rate of 2.8% each year over the past 10 years, and pediatric malignant solid tumors have become a major cause of illness and death in children. Primary surgical resection or surgical resection after other treatments is still recognized as the current first choice for the treatment of pediatric malignant solid tumors and is the only means to achieve a radical cure for malignant tumors. However, the necessity of pursuing radical surgery in children and the special characteristics of pediatric tumors put forward higher requirements for pediatric precision surgery.

4.1 Pediatric liver tumor

The anatomical structure of the liver is complex, and internal vascular and biliary tract variants are common, especially in the hepatic vein. The structure of the intrahepatic vascular system in pediatric patients is very delicate, and the organ is small in size and poorly tolerant to surgical trauma. Moreover, pediatric liver tumors are often huge, complex, fast-growing, and highly malignant. Tumors often squeeze and deform the surrounding blood vessels, and the compression or invasion of the adjacent liver area is difficult to identify. Large tumors involving the hepatic porta and tumors originating from the hepatic porta are still difficult to treat surgically. In addition, pediatric liver volume changes greatly with age and weight, so individualized liver anatomy and volume analyses are very important [11].

Hepatoblastoma (HB) is the most common primary malignant tumor of the liver in children. Its incidence rate is the highest among infants and children under 5 years old, with an annual incidence of approximately 1.5 cases per 1 million. The increasing incidence year by year and disparities between races have attracted widespread attention. With the combination of surgery and chemotherapy, especially neoadjuvant chemotherapy, the prognosis of children with HB has improved significantly, with survival rates increasing from 30% to approximately 80% [12, 13]. However, surgical resection is still an important and indispensable treatment for HB, and whether the tumor can be completely removed with a sufficient liver remnant volume is the key factor affecting the prognosis of such children [14].

All current collaborative trial groups used PRETEXT/POSTTEXT to assess the surgical resectability of HB before surgery. This staging is based on 2D cross-sectional images and is performed on the basis of Couinaud’s liver segmentation by determining the number of consecutive tumor-free liver sections. In practice, this approach is of limited help to the surgeon, and the assessment of staging and surgical resectability by window-level selection and artificial measurements is severely limited by anatomical basis, image interpretation experience, and surgical experience. Only a very rough estimate of the expected surgical procedure difficulty can be made. In addition, although Couinaud’s segmentation is very classical and practical, it is limited by the small number of dissection cases available when the classification was established and some differences between the isolated and living liver. PRETEXT also provides a detailed and cumbersome description of vascular variants, but as definitions, their clinical application is limited [15].

From the point of view of surgical resection, regardless of the strategy and staging, what must be assessed is vascular involvement, which was also defined by PRETEXT as the annotation factors V and P [13]. In clinical practice, the extent of tumor involvement in major vessels is difficult to assess due to the limitations of 2D images and the deformation variability of the liver vascular system. Another key consideration for surgical resection is the future liver remnant (FLR). An adequate postoperative FLR volume is important, as a small FLR volume can lead to acute liver failure or even death. For HB, guidelines emphasize anatomic hepatic resection, which allows for more normal liver tissue located >1 cm outside the tumor to be removed. Non-anatomic hepatic resection for advanced HB is often considered, such as extended major hepatectomies, mid-liver lobectomy, or segmental resections, which require more precise assessments of FLR [16].

3D imaging technology based on CT images is able to display the positional relationships of the liver, tumor, and all internal ductal structures in a comprehensive and simultaneous manner to achieve accurate evaluations of distances in three-dimensional space, which has obvious advantages in vessels with compression deformation or individual anatomical variations. The ability to track the route of each vessel and determine the drainage segment of each vein is important for determining individualized liver segmental anatomy [15, 17]. In addition, this technology allows for continuous assessments of preoperative chemotherapy and postoperative liver regeneration, which is of greater value in selecting the optimal timing of surgical resection and assessing postoperative liver recovery. Hisense CAS was developed based on enhanced pediatric CT data, so it has more advantages in displaying pediatric liver tumors, especially huge tumors compressing the hepatic porta. Hisense CAS can clearly show the relationships between the tumor and blood vessels and improve the resectability of liver tumors.

Two typical cases of patients with HB who underwent surgical planning with Hisense CAS are shown below. Figure 3 demonstrates a 4-year-old boy with a large liver tumor. Enhanced CT allowed for an approximate analysis of the tumor size and the adjacent relationships (Figure 3A–D). 3D imaging based on CT could show the location of the liver, the tumor, and all internal ductal structures in relation to each other in a comprehensive, whole, and simultaneous manner. When vascular involvement was evaluated after 5 cycles of neoadjuvant chemotherapy, it was found that the left hepatic vein cointersected with the middle hepatic vein and merged into the inferior vena cava. The tumor margin was only 0.5117 mm from the cointersection (Figure 3E–H). Preoperative simulation of right hemicolectomy showed that the residual liver volume was sufficient. The surgery was performed according to the preoperative plan, and the cointersection and the middle hepatic vein were successfully preserved (Figure 3I–L).

Figure 3.

Computer-assisted resection of the liver tumor with hepatic vein variation.

Figure 4 shows another 3-month-old girl with a massive tumor volume (459.1 ml) when her liver-occupying lesion was detected. The tumor was compressing and invading important blood vessels of the liver, and an aspiration biopsy confirmed HB. Neoadjuvant chemotherapy was the only option other than liver transplantation (Figure 4A and B). The tumor remained unresectable based on the evaluation performed after 4 cycles of neoadjuvant chemotherapy (Figure 4C and D). The re-evaluation after 5 cycles of neoadjuvant chemotherapy showed no significant change in tumor volume, from 35.7 ml to 35.0 ml, and the tumor was still too close to the important blood vessels of the liver and could not be operated on (Figure 4E and F). The re-evaluation after 6 cycles of neoadjuvant chemotherapy showed that the tumor was slightly reduced in size, from 35.0 ml to 25.9 ml, and the tumor was in contact with blood vessels, so surgical resection was considered (Figure 4G and H). Intraoperative 3D images assisted the surgery. The operation was successful, the tumor was completely removed, and the middle hepatic vein and portal vein were successfully preserved (Figure 4I–L).

Figure 4.

Computer-assisted resection of middle lobe tumor of the liver.

4.2 Pediatric retroperitoneal tumor

Retroperitoneal tumors (RTs) are insidious in origin, lack specific clinical manifestations, and have multiple pathological types. Due to their special anatomical location, these tumors are often found to involve large blood vessels and adjacent organs. The incidence of RTs is low, accounting for approximately 0.07%–0.20% of systemic tumors [18]. However, its malignant degree and recurrence rate are high, so complete surgical resection is the most effective treatment and affects the prognosis [19, 20]. Therefore, it is particularly important to accurately evaluate the anatomical relationships of RTs.

At present, the preoperative evaluation of RTs mainly relies on ultrasound, CT, and MRI. Among them, CT is fast, with high resolution and clear images, and can objectively reflect the compression and displacement of tumors with the surrounding organs and large blood vessels, and it has good reference value [21]. However, CT can only provide simple 2D images, and surgeons can generally judge the tumor’s size, location, and adjacent relationships by reading consecutive 2D images. However, this lacks objective accuracy and does not facilitate preoperative communication with colleagues and family members. In addition, more importantly, CT images can only show blood vessels along a specific cross-section and cannot fully display the course and wall shape of the curved blood vessels or show large compressed vessels such as the abdominal aorta, inferior vena cava, portal vein, mesenteric arteries, and iliac vessels in detail.

The application and development of digital medical technology overcame the disadvantages of CT. 3D reconstruction of CT images has made it possible to display the relationships of the tumor with surrounding adjacent organs and blood vessels in a three-dimensional, dynamic, and visualized manner. Hisense CAS can also build a 3D model, which can be rotated, scaled, and combined in any way to clearly show the size and shape of the tumor and the anatomical relationships and invasion situation between the tumor and the organs and blood vessels, especially the shape of the vasculature, thus reducing the subjective error of reading the original CT images to assess the size and degree of tumor invasion and making the preoperative assessment more realistic and reliable.

Pediatric RTs are mostly neuroblastic tumors, including neuroblastoma (NB) and ganglioneuroblastoma (GNB), which are malignant tumors, and ganglioneuroma (GN), which is a benign tumor. All three types originate from primitive neural crest cells in the neuroectoderm but are difficult to distinguish and can be mutually transformed [22]. NB is one of the most common malignant solid tumors in children and has no specific symptoms or signs. Its CT manifestations are as follows: mostly lobulated; poorly defined; often with coarse, patchy calcifications within the tumor; infiltrative growth across the midline; and high rate of involvement of the surrounding vital tissues and organs. Pediatric RTs are often found in stages III and IV and enveloping and infiltrating large retroperitoneal vessels, and up to 45% of abdominal neuroblastomas have invasion into the renal pedicle. Often the preoperative differential diagnosis between pediatric RTs and nephroblastoma becomes difficult due to excessive invasion of the kidney. This makes it difficult to resect NB. Despite chemotherapy, there are still quite a number of cases with only biopsy or partial resection, and radical surgery without a tumor at the surgical margin under the microscope is actually impossible. However, complete surgical resection of NB is the basis for further treatment, improves the confidence of both physicians and patients in treatment, and is associated with prognosis [23]. Hisense CAS assists in the anatomical analysis of important retroperitoneal vessels to improve surgical resectability and reduce the incidence of surgical complications such as vascular injury and kidney damage.

GN is generally insensitive to chemotherapy, and radical surgical resection is the first choice to confirm the diagnosis and cure. The CT manifestations are often a well-defined, regular-shaped mass, with mostly speckled calcifications in the tumor. The tumor mainly pushes and compresses the surrounding vessels and can grow along the peri-organ space and encircle the blood vessels. Despite the vessels being encircled, the vessels are generally not invaded, and the shape of the vasculature is natural and straight, with few occlusions or stenoses. To avoid sampling errors in aspiration biopsy, to relieve the symptoms of tumor compression already present, and to reduce the possibility of malignant transformation, surgical resection of suspected GN or GNB can be performed. In giant GN/GNB of retroperitoneal origin, the base of the tumor is often the mesenteric root, and involvement of the abdominal aorta, inferior vena cava, and mesenteric arteries is often the main reason for complete resection of the tumor [24]. Hisense CAS aids in the complete resection of the tumor to reduce recurrence and protects important vessels to avoid complications such as bleeding, intestinal obstruction, and intestinal necrosis.

Figure 5 shows a typical case of a 4-year-old child with an RT. Enhanced CT of the abdomen showed a huge mass-like mixed-density lesion in the abdominal cavity with a maximum cross-section of approximately 123 mm × 85 mm. The radiologists considered the mass to be a tumor (NB?), and there was a very thick blood vessel inside the tumor (Figure 5A–C). To clarify the diagnosis and decide on the next treatment, ultrasound-guided abdominal mass aspiration biopsy was performed. The pathologists first considered the mass to be a GN. Thus, surgical resection was the best option for this type of benign tumor. For precise preoperative evaluation, 3D reconstruction was performed using CAS. The reconstructed image clearly showed that the tumor was located in the retroperitoneum, and the mass had a volume of 676.7 ml. The mass was extremely close to the abdominal aorta. The superior mesenteric vein was pushed forwards, and the inferior mesenteric artery passed through the tumor (Figure 5D–F). The intraoperative exploration was completely consistent with the preoperative three-dimensional evaluation, and the tumor had a relatively complete fibrous capsule. The superior mesenteric vein was pushed to the front of the tumor. The tumor was close to the abdominal aorta, and the inferior mesenteric artery penetrated the tumor. After splitting the tumor with a CUSA knife, the inferior mesenteric artery that was encased by the tumor could be seen. Arterial pulsation was seen in the exposed inferior mesenteric artery, and the distal sigmoid colon and rectum were ruddy. The tumor section was yellowish-white with a straight and intact vascular sheath, and the postoperative tumor weight was 820 g (Figure 5G–L). The tumor was finally diagnosed as a GNB.

Figure 5.

Computer-assisted resection of retroperitoneal tumor (With the permission of the author).

4.3 Pediatric mediastinal tumor

Most mediastinal tumors have an insidious onset and lack specific clinical manifestations, and most of them have no clinical symptoms in the early stage. However, because there are many important organs and structures in the mediastinum, such as the heart, superior vena cava, trachea, and esophagus, the thorax, which has a bony structure, is not as elastic as the abdomen. Because they have less space for cushioning, mediastinal masses are prone to compressing important organs and the corresponding symptoms, namely, mediastinal mass syndrome (MMS) [25]. Compared with that in adults, the thoracic cavity in children is relatively smaller in size, and therefore, its complex anatomic-spatial relationships and dense vascular-neural structures have brought more challenges for surgical treatment [26].

Mediastinal tissues are of complex origin, and a variety of benign or malignant primary tumors can occur. Neuroblastic tumors are the most common mediastinal tumors in children. In principle, once a mediastinal mass is found, it should be actively treated. Tumors with clear borders and small volumes can be considered for radical surgery. For malignant tumors with high surgical risk, biopsy should be considered, followed by a combination of chemotherapy, surgery, radiotherapy, and immunotherapy. The difficulty of surgery lies in separating the tumor from the arteries, thoracic vertebrae, chest wall, and lung lobes. In children, the mediastinum is small, and the tissues are delicate. Thus, it is easy to inadvertently damage blood vessels and nerves during surgery, resulting in massive hemorrhage or vascular injury or even death. Therefore, accurate preoperative positioning is particularly important [27]. In addition, it has been reported that the blood vessels supplying mediastinal tumors are highly variable and may come from intercostal arteries, coronary arteries, the thyroglossal trunk, internal thoracic arteries, bronchial arteries, etc. Mediastinal tumors usually have an abundant blood supply from multiple arteries, and surgical resection may lead to severe blood loss.

Compared with 2D CT images, 3D reconstructed images can better visualize the adjacent relationships of important mediastinal tissues, whether the tumor invades the blood vessels, and the variations of the blood vessels so that surgeons can clarify the anatomical relationships. Hisense CAS aids in the precise localization of mediastinal tumors and the accurate assessment of important and variant vessels to reduce damage to vital organs and vessels [6, 17].

Figure 6 shows a three-year-old girl with a mediastinal tumor. Enhanced CT of the thorax suggested that the tumor was located in the left posterior part of the heart and T4-T9 left the paravertebral region, measuring approximately 61.15 mm × 42.00 mm. The tumor pushed the adjacent lung tissue, and part of the tumor extended to the spinal canal (Figure 6A–C). 3D reconstruction suggested that the tumor originated from the posterior mediastinum and was closely adhered to the thoracic aorta and thoracic vertebra. The three supply vessels of the tumor came from the branches of the thoracic aorta, and part of the tumor protruded into the intervertebral foramen (Figure 6D–F). Intraoperatively, the anatomical relationship of the tumor was approximately the same as the preoperative three-dimensional reconstruction results. The tumor was very densely adherent to the thoracic vertebrae and rib space, and part of the periosteum and rib space were excised to gradually remove the tumor completely (Figure 6G and H). The postoperative pathological diagnosis was a GNB.

Figure 6.

Computer-assisted resection of mediastinal tumor.

In summary, artificial intelligence technology has made significant breakthroughs and clinical applications in the field of precision surgery. Computer-assisted medical technology combines the interdisciplinary disciplines of imaging, medical image processing, and computer science, focusing on the development of assisted clinical treatment and surgical planning and simulation systems, and has become a frontier in the development of modern medical technology. Computer-assisted pediatric precise surgery improves tumor resection rates and surgical safety in a comprehensive and objective manner using artificial intelligence. In the future, individualized 3D-based precision surgery may be a new direction for surgical research.