Impact of a Categorical AI System for Digital Breast Tomosynthesis on Breast Cancer Interpretation by Both General Radiologists and Breast Imaging Specialists.

Purpose To evaluate performance improvements of general radiologists and breast imaging specialists when interpreting a set of diverse digital breast tomosynthesis (DBT) examinations with the aid of a custom-built categorical artificial intelligence (AI) system. Materials and Methods A fully balanced multireader, multicase reader study was conducted to compare the performance of 18 radiologists (nine general radiologists and nine breast imaging specialists) reading 240 retrospectively collected screening DBT mammograms (mean patient age, 59.8 years ± 11.3 [SD]; 100% women), acquired between August 2016 and March 2019, with and without the aid of a custom-built categorical AI system. The area under the receiver operating characteristic curve (AUC), sensitivity, and specificity across general radiologists and breast imaging specialists reading with versus without AI were assessed. Reader performance was also analyzed as a function of breast cancer characteristics and patient subgroups. Results Every radiologist demonstrated improved interpretation performance when reading with versus without AI, with an average AUC of 0.93 versus 0.87, demonstrating a difference in AUC of 0.06 (95% CI: 0.04, 0.08; P < .001). Improvement in AUC was observed for both general radiologists (difference of 0.08; P < .001) and breast imaging specialists (difference of 0.04; P < .001) and across all cancer characteristics (lesion type, lesion size, and pathology) and patient subgroups (race and ethnicity, age, and breast density) examined. Conclusion A categorical AI system helped improve overall radiologist interpretation performance of DBT screening mammograms for both general radiologists and breast imaging specialists and across various patient subgroups and breast cancer characteristics. Keywords: Computer-aided Diagnosis, Screening Mammography, Digital Breast Tomosynthesis, Breast Cancer, Screening, Convolutional Neural Network (CNN), Artificial Intelligence Supplemental material is available for this article. © RSNA, 2024.

Artificial intelligence (AI) to enhance breast cancer screening: protocol for population-based cohort study of cancer detection.

INTRODUCTION: Artificial intelligence (AI) algorithms for interpreting mammograms have the potential to improve the effectiveness of population breast cancer screening programmes if they can detect cancers, including interval cancers, without contributing substantially to overdiagnosis. Studies suggesting that AI has comparable or greater accuracy than radiologists commonly employ 'enriched' datasets in which cancer prevalence is higher than in population screening. Routine screening outcome metrics (cancer detection and recall rates) cannot be estimated from these datasets, and accuracy estimates may be subject to spectrum bias which limits generalisabilty to real-world screening. We aim to address these limitations by comparing the accuracy of AI and radiologists in a cohort of consecutive of women attending a real-world population breast cancer screening programme. METHODS AND ANALYSIS: A retrospective, consecutive cohort of digital mammography screens from 109 000 distinct women was assembled from BreastScreen WA (BSWA), Western Australia's biennial population screening programme, from November 2016 to December 2017. The cohort includes 761 screen-detected and 235 interval cancers. Descriptive characteristics and results of radiologist double-reading will be extracted from BSWA outcomes data collection. Mammograms will be reinterpreted by a commercial AI algorithm (DeepHealth). AI accuracy will be compared with that of radiologist single-reading based on the difference in the area under the receiver operating characteristic curve. Cancer detection and recall rates for combined AI-radiologist reading will be estimated by pairing the first radiologist read per screen with the AI algorithm, and compared with estimates for radiologist double-reading. ETHICS AND DISSEMINATION: This study has ethical approval from the Women and Newborn Health Service Ethics Committee (EC00350) and the Curtin University Human Research Ethics Committee (HRE2020-0316). Findings will be published in peer-reviewed journals and presented at national and international conferences. Results will also be disseminated to stakeholders in Australian breast cancer screening programmes and policy makers in population screening.

Robust breast cancer detection in mammography and digital breast tomosynthesis using an annotation-efficient deep learning approach.

Breast cancer remains a global challenge, causing over 600,000 deaths in 2018 (ref. 1). To achieve earlier cancer detection, health organizations worldwide recommend screening mammography, which is estimated to decrease breast cancer mortality by 20-40% (refs. 2,3). Despite the clear value of screening mammography, significant false positive and false negative rates along with non-uniformities in expert reader availability leave opportunities for improving quality and access4,5. To address these limitations, there has been much recent interest in applying deep learning to mammography6-18, and these efforts have highlighted two key difficulties: obtaining large amounts of annotated training data and ensuring generalization across populations, acquisition equipment and modalities. Here we present an annotation-efficient deep learning approach that (1) achieves state-of-the-art performance in mammogram classification, (2) successfully extends to digital breast tomosynthesis (DBT; '3D mammography'), (3) detects cancers in clinically negative prior mammograms of patients with cancer, (4) generalizes well to a population with low screening rates and (5) outperforms five out of five full-time breast-imaging specialists with an average increase in sensitivity of 14%. By creating new 'maximum suspicion projection' (MSP) images from DBT data, our progressively trained, multiple-instance learning approach effectively trains on DBT exams using only breast-level labels while maintaining localization-based interpretability. Altogether, our results demonstrate promise towards software that can improve the accuracy of and access to screening mammography worldwide.

Functions of ventral visual cortex after bilateral medial temporal lobe damage.

Repeated stimuli elicit attenuated responses in visual cortex relative to novel stimuli. This adaptation can be considered as a form of rapid learning and a signature of perceptual memory. Adaptation occurs not only when a stimulus is repeated immediately, but also when there is a lag in terms of time and other intervening stimuli before the repetition. But how does the visual system keep track of which stimuli are repeated, especially after long delays and many intervening stimuli? We hypothesized that the hippocampus and medial temporal lobe (MTL) support long-lag adaptation, given that this memory system can learn from single experiences, maintain information over delays, and send feedback to visual cortex. We tested this hypothesis with fMRI in an amnesic patient, LSJ, who has encephalitic damage to the MTL resulting in extensive bilateral lesions including complete hippocampal loss. We measured adaptation at varying time lags between repetitions in functionally localized visual areas that were intact in LSJ. We observed that these areas track information over a few minutes even when the hippocampus and extended parts of the MTL are unavailable. LSJ and controls were identical when attention was directed away from the repeating stimuli: adaptation occurred for lags up to three minutes, but not six minutes. However, when attention was directed toward stimuli, controls now showed an adaptation effect at six minutes but LSJ did not. These findings suggest that visual cortex can support one-shot perceptual memories lasting for several minutes but that the hippocampus and surrounding MTL structures are necessary for adaptation in visual cortex after longer delays when stimuli are task-relevant.

A Neural Basis for Developmental Topographic Disorientation.

Developmental topographic disorientation (DTD) is a life-long condition in which affected individuals are severely impaired in navigating around their environment. Individuals with DTD have no apparent structural brain damage on conventional imaging and the neural mechanisms underlying DTD are currently unknown. Using functional and diffusion tensor imaging, we present a comprehensive neuroimaging study of an individual, J.N., with well defined DTD. J.N. has intact scene-selective responses in the parahippocampal place area (PPA), transverse occipital sulcus, and retrosplenial cortex (RSC), key regions associated with scene perception and navigation. However, detailed fMRI studies probing selective tuning properties of these regions, as well as functional connectivity, suggest that J.N.'s RSC has an atypical response profile and an atypical functional coupling to PPA compared with human controls. This deviant functional profile of RSC is not due to compromised structural connectivity. This comprehensive examination suggests that the RSC may play a key role in navigation-related processing and that an alteration of the RSC's functional properties may serve as the neural basis for DTD. SIGNIFICANCE STATEMENT: Individuals with developmental topographic disorientation (DTD) have a life-long impairment in spatial navigation in the absence of brain damage, neurological conditions, or basic perceptual or memory deficits. Although progress has been made in identifying brain regions that subserve normal navigation, the neural basis of DTD is unknown. Using functional and structural neuroimaging and detailed statistical analyses, we investigated the brain regions typically involved in navigation and scene processing in a representative DTD individual, J.N. Although scene-selective regions were identified, closer scrutiny indicated that these areas, specifically the retrosplenial cortex (RSC), were functionally disrupted in J.N. This comprehensive examination of a representative DTD individual provides insight into the neural basis of DTD and the role of the RSC in navigation-related processing.

Attention flexibly alters tuning for object categories.

Using functional MRI (fMRI) and a sophisticated forward encoding and decoding approach across the cortical surface, a new study examines how attention alternates tuning functions across a large set of semantic categories. The results suggest a dynamic attention mechanism that allocates greater resources to the attended and related semantic categories at the expense of unattended ones.

Greater sensitivity to nonaccidental than metric changes in the relations between simple shapes in the lateral occipital cortex.

Behavioral studies and single cell recordings in monkey inferotemporal cortex have documented greater sensitivity to differences in viewpoint invariant or nonaccidental properties (e.g., straight vs. curved), than metric properties (e.g., degree of curvature) of simple shapes. Are we similarly more sensitive to nonaccidental (NAP) than metric (MP) differences of the relations between objects? We addressed this question with sets of scene triplets that could, from a reference or "Base" scene (e.g., a brick slightly separated from a cylinder), undergo a NAP relational change (e.g., the brick attached to the cylinder) or an MP relational change (e.g., the brick further separated from the cylinder). Critically, both relational variations were matched in physical dissimilarity using pixel energy and the Gabor-jet system, a model of V1 similarity. In an adaptive staircase match-to-sample paradigm, subjects required more than double the presentation durations for detecting differences in MP than NAP relations to achieve equivalent levels of accuracy. In two fMRI experiments, NAP changes consistently produced greater responses in the lateral occipital cortex (LO), but not in earlier retinotopic stages, compared to MP changes, implicating LO as the potential neural locus for where the greater detectability of the differences of NAPs than MPs is made explicit. HMAX, a model of cell tuning in higher-level ventral visual areas, did not consistently reflect the marked NAP advantage witnessed in behavioral performance and in LO responses.

The benefit of object interactions arises in the lateral occipital cortex independent of attentional modulation from the intraparietal sulcus: a transcranial magnetic stimulation study.

Our visual experience is generally not of isolated objects, but of scenes, where multiple objects are interacting. Such interactions (e.g., a watering can positioned to pour water toward a plant) have been shown to facilitate object identification compared with when the objects are depicted as not interacting (e.g., a watering can positioned away from the plant) (Green and Hummel, 2004, 2006). What is the neural basis for this advantage? Recent fMRI studies have identified the lateral occipital cortex (LO) as a potential neural origin of this behavioral benefit, as LO showed greater responses to object pairs depicted as interacting compared with when they are not (Kim and Biederman, 2010; Roberts and Humphreys, 2010). However, it is possible that LO was modulated by an attention-sensitive region, the intraparietal sulcus (IPS), which sometimes showed a similar pattern of responses as that of LO in the Kim and Biederman (2010) investigation. To test this hypothesis, we delivered transcranial magnetic stimulation (TMS) to human subjects' LO and IPS while they detected a target object that was or was not interacting with another object to form a scene. TMS delivered to LO but not IPS abolished the facilitation in identifying interacting objects compared with noninteracting depictions observed in the absence of TMS, suggesting that it is LO and not IPS that is critical for the coding of object interactions.

Where do objects become scenes?

Regions tuned to individual visual categories, such as faces and objects, have been discovered in the later stages of the ventral visual pathway in the cortex. But most visual experience is composed of scenes, where multiple objects are interacting. Such interactions are readily described by prepositions or verb forms, for example, a bird perched on a birdhouse. At what stage in the pathway does sensitivity to such interactions arise? Here we report that object pairs shown as interacting, compared with their side-by-side depiction (e.g., a bird besides a birdhouse), elicit greater activity in the lateral occipital complex, the earliest cortical region where shape is distinguished from texture. Novelty of the interactions magnified this gain, an effect that was absent in the side-by-side depictions. Scene-like relations are thus likely achieved simultaneously with the specification of object shape.

Adaptation in the fusiform face area (FFA): image or person?

Viewing a sequence of faces of two different people results in a greater Blood Oxygenation Level Dependent (BOLD) response in FFA compared to a sequence of identical faces. Changes in identity, however, necessarily involve changes in the image. Is the release from adaptation a result of a change in face identity, per se, or could it be an effect that would arise from any change in the image of a face? Subjects viewed a sequence of two faces that could be of the same or different person, and in the same or different orientation in depth. Critically, the physical similarity of view changes of the same person was scaled, by Gabor-jet differences, to be equivalent to that produced by an identity change. Both person and orientation changes produced equivalent releases from adaptation in FFA (relative to identical faces) suggesting that FFA is sensitive to the physical similarity of faces rather than to the individuals depicted in the images.

Adaptation to objects in the lateral occipital complex (LOC): shape or semantics?

A change in the basic-level class when viewing a sequence of two objects produces a large release from adaptation in LOC compared to when the images are identical. Is this due to a change in semantics or shape? In an fMRI-adaptation experiment, subjects viewed a sequence of two objects and judged whether the stimuli were identical in shape. Different-shaped stimuli could be from the same or different basic-level classes, where the physical similarities of the pairs in the two conditions were equated by a model of simple cell similarity. BOLD responses in LOC for the two conditions were equivalent, and higher than that of the identical condition, indicating that LOC is sensitive to shape rather than to basic-level semantics.

Blind or deaf? A matter of aesthetics.

Blind or deaf? Surprisingly, a small but noticeable minority, 17%, opted for blindness. Another 6% were indifferent. For these individuals (23% of the sample), the loss of the aesthetic experience--music--was given relatively greater weight than the loss of the practical--conversation. This was not a symmetric effect in that those who opted for being deaf did not place greater value on the aesthetic aspects of vision compared to those who opted for being blind. What predicts the preference for being blind was not the amount of time spent listening to music but the extent to which one formally studied, created, and played music and experienced intense emotions when engaged by music.

17,000 years of depicting the junction of two smooth shapes.

Competent realistic drawings preserve viewpoint-invariant shape characteristics of simple parts, such that a contour in the object that is straight or curved, for example, is depicted that way in the drawing. A more subtle invariant--a V-shaped singularity of the occluding boundary, containing a T-junction and a contour termination--is produced at the junction between articulated smooth surfaces, as with the leg joining the body of a horse. 45% of the drawings made in 2007 by individuals with only minimal art education correctly depicted such junctions, a proportion that is not reliably different from the incidence (42%) of correct depictions in a large sample of cave art made 17000 years ago. Whether a person did or did not include the invariant in their drawing, all agreed that it made for a better depiction.