AnomalousNet: a hybrid approach with attention U-Nets and change point detection for accurate characterization of anomalous diffusion in video data

Abstract Anomalous diffusion is ubiquitous in systems ranging from intracellular transport and porous-medium flow to animal foraging, and its quantification requires robust methods that cope with short and noisy trajectories. We present AnomalousNet, a unified three-stage pipeline for analyzing anomalous particle dynamics. First, up to 64 particles per <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:mn>128</mml:mn> <mml:mo>×</mml:mo> <mml:mn>128</mml:mn> </mml:mrow> </mml:math> field of view are tracked using the Crocker–Grier algorithm via Trackpy . Next, an Attention U-Net trained on over 8.6 × 10 4 simulated experiments infers frame-wise anomalous exponents α , generalized diffusion coefficients K , and discrete motion states. Finally, regime changes are identified using an L2-regularized, windowed pruned exact linear time change-point detection algorithm. On the 2nd Anomalous Diffusion Challenge benchmark the method achieved <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:mtext>MAE</mml:mtext> <mml:mo stretchy="false">(</mml:mo> <mml:mi>α</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>=</mml:mo> <mml:mn>0.32</mml:mn> </mml:mrow> </mml:math> , <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:mtext>MSLE</mml:mtext> <mml:mo stretchy="false">(</mml:mo> <mml:mi>K</mml:mi> <mml:mo stretchy="false">)</mml:mo> <mml:mo>=</mml:mo> <mml:mn>0.10</mml:mn> </mml:mrow> </mml:math> , state-classification <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:msub> <mml:mi>F</mml:mi> <mml:mrow> <mml:mn>1</mml:mn> </mml:mrow> </mml:msub> <mml:mo>=</mml:mo> <mml:mn>0.93</mml:mn> </mml:mrow> </mml:math> , and change-point RMSE = 0.09, ranking second in the video single-trajectory task and ensemble task. These results demonstrate precise discrimination of subdiffusive, normal, and superdiffusive regimes, and frame-level identification of state transitions establishing AnomalousNet as a powerful tool for quantitative analysis of heterogeneous diffusion in video data.