Surapa Thiemjarus

A probabilistic framework for tracking deformable soft tissue in minimally invasive surgery

The use of vision based algorithms in minimally invasive surgery has attracted significant attention in recent years due to its potential in providing in situ 3D tissue deformation recovery for intra-operative surgical guidance and robotic navigation. Thus far, a large number of feature descriptors have been proposed in computer vision but direct application of these techniques to minimally invasive surgery has shown significant problems due to free-form tissue deformation and varying visual appearances of surgical scenes. This paper evaluates the current state-of-the-art feature descriptors in computer vision and outlines their respective performance issues when used for deformation tracking. A novel probabilistic framework for selecting the most discriminative descriptors is presented and a Bayesian fusion method is used to boost the accuracy and temporal persistency of soft-tissue deformation tracking. The performance of the proposed method is evaluated with both simulated data with known ground truth, as well as in vivo video sequences recorded from robotic assisted MIS procedures.

Multi-sensor Fusion

In the previous chapters, we have discussed issues concerning hardware, communication and network topologies for the practical deployment of Body Sensor Networks (BSNs). The pursuit of low power miniaturised distributed sensing under a patient’s natural physiological conditions has also imposed significant technical challenges on integrating information from what is often heterogeneous, incomplete and error-prone sensor data. For BSNs, the nature of errors can be attributed to a number of sources; but motion artefacts, inherent limitations and possible malfunctions of the sensors along with communication errors are the main causes of concern. In practice, it is desirable to rely on sensors with redundant or complementary data to maximise the information content and reduce both systematic errors and random artefacts. This, in essence, is the main drive for multi-sensor fusion, which is concerned with the synergistic use of multiple sources of information.

A spatio-temporal architecture for context aware sensing

Context-aware sensing is an integral part of the body sensor network (BSN) design and it allows the understanding of intrinsic characteristics of the sensed signal and determination of how BSNs should react to different events and adapt its monitoring behaviour. The purpose of this paper is to propose a novel spatio-temporal self-organising map that minimises the number of neurons involved whilst maintaining a high accuracy in class separation for both static and dynamic activities

A study on instance-based learning with reduced training prototypes for device-context-independent activity recognition on a mobile phone

This paper presents a study of two simple methods for reducing the complexity of the instance-based classification technique and demonstrates their use in device-context independent activity recognition on a mobile phone. A projection-based method for signal rectification has been implemented on an iPhone in order to handle with variation in device orientations. The transformation matrix is estimated on a ten-second dynamic data buffer. To search for a suitable set of training prototypes for iPhone implementation, an activity recognition experiment is conducted with twenty different device contexts performed by eight subjects. With the developed mobile application, the recognition results along with the users location can be displayed on both iPhone and the web application in real time.

Adaptive Bayesian networks for video processing

Due to its static nature, the inference capability of Bayesian networks (BNs) often deteriorates when the basis of input data varies, especially in video processing applications where the environment often changes constantly. This paper presents an adaptive BN where the network parameters are adjusted in accordance to input variations. An efficient retraining method is introduced for updating the parameters and the proposed network is applied to shadow removal in video sequence processing with quantitative results demonstrating the significance of adapting the network with environmental changes.

Probabilistic decision level fusion for real-time correlation of ambient and wearable sensors

Fusing data from ambient and wearable sensors when performing in-home healthcare monitoring allows for high accuracy activity inference due to the complementary nature of sensing modalities. Where residences may house multiple occupants, we must automatically identify related data streams before fusion may occur, a process known as sensor correlation. In this paper a multi-objective variant of the Bayesian Framework for Feature Selection (BFFS) is used to construct small inter-sensor redundant feature sets which train efficient per-sensor activity classifiers. Probabilistic decision level fusion is then used to deal with noisy and erroneous sensor data and perform real-time correlation. The potential value of the proposed algorithm for pervasive sensing is demonstrated with both simulated and experimental data.

Context Aware Sensing

In recent years, there have been considerable interests in context-aware sensing for pervasive computing. Context can be defined as “the circumstances in which an event occurs” and this concept has been successfully used in information processing for over 50 years, particularly for Natural Language Processing (NLP) and Human Computer Interaction (HCI). The popularity of the context-aware architectures is due to the increasingly ubiquitous nature of the sensors, as well as the diversity of the environment under which the sensed signals are collected. To understand the intrinsic characteristics of the sensed signals and determine how BSNs should react to different events, the contextual information is essential to the adaptation of the monitoring device so as to provide more intelligent support to the users.

Toward a mixed-signal reconfigurable ASIC for real-time activity recognition

In recent years, there have been increasing interests in context aware sensing based upon ultra-low power wearable sensors. These applications require efficient processing-on-node capabilities to minimise the overall power consumption and wireless transmission bandwidths. In this paper, a novel reconfigurable mixed-signal ASIC designed for real-time activity recognition has been proposed. The system architecture integrates all signal conditioning and data processing circuits onto a single silicon substrate with configurable analogue computing and artificial neuron network-inspired classification blocks. The ASIC is designed using conventional EDA tools and has been fabricated using AMS 0.35mum CMOS technology with a final chip size of 23.8 mm2. An on-chip inferencing engine derived from off-chip training data has been developed. Both design considerations and implementation details of the ASIC are discussed. Preliminary simulation results indicate the desired performance of the ASIC for real-time activity classification.

Optimal Feature Selection Applied to Multispectral Fluorescence Imaging

Recent rapid developments in multi-modal optical imaging have created a significant clinical demand for its in vivo--in situ application. This offers the potential for real-time tissue characterization, functional assessment, and intra-operative guidance. One of the key requirements for in vivo consideration is to minimise the acquisition window to avoid tissue motion and deformation, whilst making the best use of the available photons to account for correlation or redundancy between different dimensions. The purpose of this paper is to propose a feature selection framework to identify the best combination of features for discriminating between different tissue classes such that redundant or irrelevant information can be avoided during data acquisition. The method is based on a Bayesian framework for feature selection by using the receiver operating characteristic curves to determine the most pertinent data to capture. This represents a general technique that can be applied to different multi-modal imaging modalities and initial results derived from phantom and ex vivo tissue experiments demonstrate the potential clinical value of the technique.

An integrated inferencing framework for context sensing

This paper presents the use of distributed inferencing with resource optimisation and Spatio-Temporal Self-Organising Map (STSOM) for effectively combining the wearable and ambient sensors. STSOM is an efficient local processing technique which is also suitable for enhancing the temporal behaviour of the distributed inferencing model. To reduce the complexity of the distributed model, a multi-objective Bayesian framework for feature selection has been proposed for model learning. The validation of the techniques has been conducted with activity recognition with both wearable and ambient sensors in a lab-based home monitoring setting.