The goal of interactive search-assisted diagnosis (ISAD) is to enable doctors to make more informed decisions about a given case by providing a selection of similar annotated cases. For instance, a radiologist examining a suspicious mass could study labeled mammograms with similar conditions and weigh the outcome of their biopsy results before determining whether to recommend a biopsy. The fundamental challenge in developing ISAD systems is the identification of similar cases, not simply in terms of superficial image characteristics, but in a medically-relevant sense. This task involves three aspects: extraction of a representative set of features, identifying an appropriate measure of similarity in the high-dimensional feature space, and return the most similar matches at interactive speed. The first has been an active research area for several decades. The second has largely been ignored by the medical imaging community. The third can be achieved using the Diamond framework, an open-source platform that enables efficient exploration of large distributed complex data repositories. This paper focuses on the second aspect. We show that the choice of distance metric affects the accuracy of an ISAD system and that machine learning enables the construction of effective domain-specific distance metrics. In the learned distance, data points with the same labels (e.g., malignant masses) are closer than data points with different labels (e.g., malignant vs. benign). Thus, the labels of the near neighbors of a new case are likely to be informative. We present and evaluate several novel methods for distance metric learning and evaluate them on a database involving 2522 mass regions of interest (ROI) extracted from digital mammograms, with ground truth defined by biopsy results (1800 malignant, 722 benign). Our results show that learned distance metrics improve both classification (ROC curve) and retrieval performance.
In independent vehicle concepts for the Automated Highway System (AHS), the ability to make competent tactical-level decisions in real-time is crucial. Traditional approaches to tactical reasoning typically involve the implementation of large monolithic systems, such as decision trees or finite state machines. However, as the complexity of the environment grows, the unforeseen interactions between components can make modifications to such systems very challenging. For example, changing an overtaking behavior may require several, non-local changes to car-following, lane changing and gap acceptance rules. This paper presents a distributed solution to the problem. PolySAPIENT consists of a collection of autonomous modules, each specializing in a particular aspect of the driving task - classified by traffic entities rather than tactical behavior. Thus, the influence of the vehicle ahead on the available actions is managed by one reasoning object, while the implications of an approaching exit are managed by another. The independent recommendations form these reasoning objects are expressed in the form of votes and vetos over a 'tactical action space', and are resolved by a voting arbiter. This local independence enables PolySAPIENT reasoning objects to be developed independently, using a heterogenous implementation. PolySAPIENT vehicles are implemented in the SHIVA tactical highway simulator, whose vehicles are based on the Carnegie Mellon Navlab robots.
Sensor technology plays a critical role in the operation of the Automated Highway System (AHS). The proposed concepts depend on a variety of sensors for positioning, lane- tracking, range and vehicle proximity. Since large substations of the AHS will be designed and evaluated in simulation before deployment, it is important that simulators make realistic sensor assumptions. Unfortunately, the current physical sensor models are inadequate for this task since they require detailed world state information that is unavailable in a simulated environment. In this paper, we present an open-ended, functional sensor hierarchy, incorporating geometric models and abstract noise characteristics, which can be used directly with current AHS tools. These models capture the aspects of sensing technology that are important to AHS concept design such as occlusion, and field of view restrictions, while ignoring physical-level details such as electromagnetic sensor reflections. Since the functional sensor models operate at the same level of granularity as the simulation platform, complete integration is assured. The hierarchy classifies sensors into functional groups. The models at a particular level incorporate characteristics that are common to all sensors in its subgroups. For example, range sensors have a parameter corresponding to a maximum effective range, while lane-trackers include information pertaining to lateral accuracy.