Low-cost 3D Human Body Reconstruction System for Medical Purpose

In the last decade, 3D modeling industry enjoyed booming development in both hardware and software. For hardware, wide variety of sensors were invented, some of which are relative high-end, costing thousands dollar with resolution level in the sub-millimeter range; while others are cheap but limited in both resolution and data quality. For software, most of the 3D reconstruction systems using low cost hardware aims at creating avatars for gaming, animation, 3D printing or fashion industry, but have low capabilities in reconstruction accuracy and reliability. On the other hand, there is no easy access for public to high-level medical grade scan system to scan themselves and get feedback with their health data, because systems designed for medical institutions are typically expensive, immobile and require trained professionals to operate. Moreover, most of the 3D shape reconstruction systems designed for medical purposes have rigorous pose requirements, such as having the subjects stand still and hold a pose for minutes. We present a cost-efficient easy to use 3D human body scan system for medical applications using entertainment level depth sensors which provide reconstruction result with high degree of accuracy and reliability. Users are able to scan themselves in a flexible way, rotating roughly 45°each time and hold their pose for 1-2 seconds, instead of being required to stand still on a turntable or hold one pose for several minutes. Our system enables the public to get easy access to their health data relative to their shapes, such as to estimate their body fat or to track the changes of body shape during a certain period of time.

Participants: Yao Lu, Shang Zhao, James Hahn, Geoffrey Hudson, Melissa Napolitano, Jerry Danoff

Computer-based Quantitative Assessment of Learner Performance during Neonatal Endotracheal Intubation: A Pilot Study

Success of neonatal endotracheal intubation (ETI) is measured by the ability of the operator to place the endotracheal tube in the correct position within 30 seconds. Although it is a critical skill, pediatric trainees often are unsuccessful at its performance. Task trainers and animal models coupled with instructor feedback have been traditionally used to teach neonatal intubation to residents. The small size of the neonatal airway and the large class size and limited time available during training sessions, however, hinder instructors' ability to detect specific causes of procedural failure that can be

used as feedback to direct individual learning. Smart tools that enable rapid detection of learner deficits can enhance the learning process by facilitating immediate feedback. The objectives of this pilot study were to: 1. develop a computer based quantitative assessment tool capable of recording and measuring learner performance during simulated neonatal ETI; and 2. compare novice ETI performance to expert performance. A standard neonatal resuscitation mannequin head, laryngoscope and 3.0 endotracheal tube (ETT) were fitted with electromagnetic trackers to capture mannequin head motion and the motion of the laryngoscope and ETT with 6 degrees of freedom. Replica 3D computer models of the head, laryngoscope and endotracheal tube were then developed and registered to align completely with their physical counterparts. All motions were captured and mirrored by the 3D model.

Following a warm up period, participants were recorded performing endotracheal intubation three times. Participants recruited included expert neonatology attendings (more-than 60 patient intubations), novice nurse practitioners and novice pediatric residents (less-than 25 intubations). Data was processed and simultaneously sent to a laptop screen for continuous, real-time display of the mannequin, laryngoscope and ETT position and orientation. The software recorded each procedure allowing later review by the instructor.

Participants: Lamia Soghier MD, Wei Li, Rehab Alahmadi, Randall Burd MD, James Hahn
[Poster] [Demo Video]

Generalized Temporal Focus+Context Framework for Improved Medical Data Exploration

Physicians use slices and 3D volume visualizations to place a diagnosis, establish a treatment plan and as a guide during surgical procedures. There is an observed difference in 2D and 3D visualization objectives of the various groups of specialists. We describe a generalized temporal focus+context framework that unifies different widely used and novel visualization methods. The framework is used to classify already existing common techniques and to define new techniques that can be used in medical volume visualization.

The new techniques explore the time-dependent position of the framework focus region to combine 2D and 3D rendering inside the focus and to provide a new focus-driven context region that gives explicit spatial perception cues between the current and past regions of interest. An arbitrary-shaped focus region and no context rendering are two novel framework-based techniques that support improved planning of procedures that involve drilling or endoscopic exploration. The new techniques are quantitatively compared to already existing techniques by means of a user study.

Participants: Nadezhda Radeva, Lucien Levy, James Hahn
[Published Paper]

Computer-Based Planning System for Mandibular Reconstruction Surgery

Tumors that involve the oral cavity, can arise from the tongue, floor of the mouth, lips, and hard palate, among other structures. Advanced tumors can be quite destructive and invade the mandible (lower jaw) or maxilla (upper jaw). Affected segments of the mandible or maxilla are then removed surgically. The standard of care for bony reconstruction of mandibular and maxillary defects involves use of the fibula free flap. The fibula bone is harvested and brought up into the neck with its attached pedicle, the peroneal artery and vein. The peroneal artery and vein are then anastomosed to donor

arteries and veins in the neck and the reconstructed bone segment is plated into place. Dental implants can then be placed into the reconstructed mandible at a later time to provide for dental rehabilitation. Depending on the type of defect that is created, wedges of the fibula may need to be osteotomized in the horizontal and/or vertical dimensions to re-create the normal shape of the native patient's mandible. Additionally, skin overlying the lateral aspect of the leg is also harvested along with the fibula if needed. This skin is supplied by 2-3 sub-millimeter perforator vessels coming off the peroneal artery and vein just posterior to the fibula at different points along the length of the peroneal vessels. For these composite defects, osteotomies need to be performed without damaging the perforator vessels to the skin, which are usually not visualized well during the dissection.

Ideally the wedge osteotomies are performed in a segment of the fibula bone which is away from the perforator vessels in order to prevent inadvertent injury. Historically after mandibulectomy, for large defects requiring fibular free flaps, the reconstructive surgeon created a mental three-dimensional (3D) picture in his/her mind, and reconstructed the defect without the use of pre-operative templates or guides. Pre-operative planning is crucial to this kind of surgery. Optimizing the surgery parameters is also needed.

Participants: Manal Aassaf, Wei Li, Arjun Joshi, James Hahn
[Published Paper]

Non-rigid Surface Registration using Cover Tree based Clustering and Nearest Neighbor Search

We propose a novel non-rigid registration method that computes the correspondences of two deformable surfaces using the cover tree. The aim is to find the correct correspondences without landmark selection and to reduce the computational complexity. The source surface S is initially aligned to the target surface T to generate a cover tree from the densely distributed surface points. The cover tree is constructed by taking into account the positions and normal vectors of the points and used for hierarchical clustering and nearest neighbor search. The cover tree based clustering divides the two surfaces into several clusters based on the geometric features, and each cluster on the source surface is transformed to its corresponding cluster on the target. The nearestneighbor search from the cover tree reduces the search space for correspondence computation, and the source surface is deformed to the target by optimizing the point pairs. The correct correspondence of a given source point is determined by choosing one target point with the best correspondence measure from the k nearest neighbors. The proposed energy function with Jacobian penalty allows deforming the surface accurately and with less deformation folding.

Participants: Manal Aassaf, Yeny Yim, James Hahn
[Published Paper]
Go to Past Research Page