Answers are grounded in this paper and cite their sources.
Depression recognition based on functional near-infrared spectroscopy (fNIRS) has recently been a research hotspot to explore brain activity and mental disease mechanisms. However, most existing studies ignore the task-stage structure of fNIRS data, despite its strong correlation with task-evoked brain responses. This paper proposes a Dual-stream Task-Aware model via inheritable Prompt learning (DuTAP) to establish the relationship between segmented data and task stage for fNIRS-based depression recognition. In the DuTAP, one stream encodes the entire sequence and extracts global feature, and the other stream respectively encodes segmented sequences divided by the fixed task stage slots and extracts task features via a novel T-Former module. To fully leverage task-related information, we introduce a learnable prompt for each segment and design an inheritable prompt mechanism that transmits contextual information across stages to enhance the temporal information. Two contrastive losses are employed to optimize both streams: one captures inter-sample discriminability at the global level, while the other promotes intra-sample diversity across task segments. The proposed method achieves accuracy of 89.7% on the fNIRS96 dataset and a state-of-the-art accuracy of 93.2% on an extended fNIRS1110 dataset, demonstrating the effectiveness of DuTAP. We will publicly release the fNIRS datasets and code at link.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: