Microsaccade Dynamics Enhance Nodule Detection and Discriminate Reader Expertise Levels During Chest Radiography

Technical Summary: Microsaccade Dynamics Enhance Nodule Detection and Discriminate Reader Expertise Levels During Chest Radiography

Problem Statement
Perceptual misses constitute a primary source of diagnostic error in radiology, often terminating the diagnostic pathway prematurely and risking patient harm. While radiologists develop perceptual skills to interpret chest radiographs, the physiological mechanisms underlying successful abnormality detection remain poorly understood. Furthermore, while eye-tracking variables (e.g., fixation allocation, dwell time, scanpath structure) have been used to differentiate expert from novice performance, these measures rely on broader gaze behaviors that are susceptible to voluntary cognitive strategies and bias. There is currently no known physiological biomarker for expertise in chest radiograph interpretation that is rooted in non-volitional oculomotor processes. This study addresses the gap in understanding how fixational eye movements, specifically microsaccades, contribute to pulmonary nodule detection and whether their dynamics can serve as an objective, largely non-volitional signature of reader expertise.

Methodology
The study employed an experimental eye-tracking design involving 26 adult participants categorized into three groups based on performance accuracy rather than academic rank: "low accuracy" (naive observers, N=9), "intermediate accuracy" (radiology residents, N=5), and "high accuracy" (attending radiologists and high-performing residents, N=12). This re-categorization was necessary because years of training do not always correlate linearly with perceptual performance.

• Stimuli: Participants viewed 148 chest radiographs from the Japanese Society of Radiological Technology (JSRT) database, each containing a single lung nodule. Images were ranked by subtlety (Rank 1: extremely subtle to Rank 5: obvious).
• Task: Participants searched for the nodule and indicated its location by pressing a button on a gamepad while maintaining gaze on the target. A response was deemed correct if the gaze was within a 1.5° radius target window at the moment of the button press.
• Data Acquisition: Eye movements were recorded binocularly at 500 Hz using an EyeLink 1000.
• Analysis:
  • Microsaccade Definition: Saccades with a magnitude <1° were identified using a modified Engbert and Kliegl algorithm.
  • Comparisons: Microsaccade dynamics (rate, prevalence per fixation, magnitude, duration) were compared inside versus outside the nodule target window and between correct versus incorrect trials.
  • Duration Matching: To rule out the confound of longer fixation times, analyses were repeated on duration-matched fixations (1:1 correspondence between inside and outside target windows).
  • Expertise Discrimination: Receiver Operating Characteristic (ROC) analyses were performed to determine if oculomotor variables could discriminate low-accuracy readers from intermediate/high-accuracy readers. Significance was assessed via permutation testing (n=5000).

Key Results

1. Microsaccades and Detection Success: On correct trials, across all expertise levels, readers generated significantly more microsaccades per fixation when viewing the nodule region compared to other image regions. This relationship was absent on incorrect trials, linking microsaccade production directly to successful abnormality detection.
2. Independence from Fixation Duration: The increased prevalence of microsaccades inside the target window persisted even when fixation durations were matched between inside and outside the target. This indicates that the effect is not solely driven by longer viewing times but represents a specific oculomotor strategy associated with scrutiny of the target.
3. Expertise Signatures:
   • Total Count: There was no significant difference in the total number of microsaccades produced across accuracy groups.
   • Targeted Production: Higher-accuracy readers showed a greater prevalence of fixations containing microsaccades specifically within the target region on correct trials.
   • ROC Analysis: ROC analyses identified a microsaccade-based oculomotor signature capable of discriminating lower- from higher-accuracy readers. Specifically, the microsaccade rate within the target region on correct trials served as a significant discriminator (AUC = 0.82, p < 0.005).
4. Search Behavior: High-accuracy participants spent more time searching on difficult trials but less time on obvious trials compared to low-accuracy participants. On incorrect trials, high-accuracy participants searched longer and foveated a larger portion of the image than low-accuracy participants, suggesting persistence driven by confidence in the presence of a nodule.

Key Contributions

• Functional Role of Microsaccades: The study provides evidence that microsaccade dynamics enhance pulmonary nodule detection. The production of microsaccades during the foveation of a target is a functional component of successful detection, distinct from mere fixation duration.
• Physiological Biomarker of Expertise: The paper identifies a microsaccade-based oculomotor signature that differentiates expert from novice performance. Unlike broader gaze measures (e.g., scanpath structure) which are subject to voluntary strategy, microsaccades are largely non-volitional and occur outside conscious awareness.
• Methodological Refinement: By decoupling microsaccade production from fixation duration through duration-matched analyses, the study isolates microsaccades as an independent variable in visual search performance.

Significance and Claims
The authors claim that this work provides, to their knowledge, the first microsaccade-based, largely non-volitional physiological biomarker of expertise in chest radiograph interpretation.

The significance of these findings lies in the nature of the signal: because microsaccades are robust to voluntary modulation and occur outside conscious awareness, they offer an objective readout of perceptual expertise that is less susceptible to the strategic biases inherent in overt gaze behaviors. The study suggests that accurate medical image interpretation depends on generating microsaccades within a critical spatial and temporal window. Consequently, these findings open a new avenue for studying and potentially assessing radiological expertise through a physiological signal that reflects stable differences in non-volitional oculomotor processing shaped by training and experience. The authors do not claim that microsaccades are the sole determinant of expertise, but rather that they are a functional component of successful detection and a reliable discriminator of reader performance levels.