Tianhang Nan

Tianhang Nan

Auto-generated profile, based on publicly available information

Tianhang Nan

College of Medicine and Biological Information Engineering, Northeastern University, Shenyang, China

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
8 April 2024 in IEEE Journal of Biomedical and Health Informatics
Verified
Journal Article
Closed access
HQHao Quan
Hao Quan
XLXinjia Li
Xinjia Li
DHDayu Hu
Dayu Hu
TNTianhang Nan
Tianhang Nan
XCXiaoyu Cui
Xiaoyu Cui
In the field of pathology, the scarcity of certain diseases and the difficulty of annotating images hinder the development of large, high-quality datasets, which in turn affects the advancement of deep learning-assisted diagnostics. Few-shot learning has demonstrated unique advantages in modeling tasks with limited data, yet explorations of this method in the field of pathology remain in the early stages. To address this issue, we present a dual-channel prototype network (DCPN), a novel few-shot learning approach for efficiently classifying pathology images with limited data. The DCPN leverages self-supervised learning to extend the pyramid vision transformer (PVT) to few-shot classification tasks and combines it with a convolutional neural network to construct a dual-channel network for extracting multi-scale, high-precision pathological features, thereby substantially enhancing the generalizability of prototype representations. Additionally, we design a soft voting classifier based on multi-scale features to further augment the discriminative power of the model in complex pathology image classification tasks. We constructed three few-shot classification tasks with varying degrees of domain shift using three publicly available pathological datasets—CRCTP, NCTCRC, and LC25000—to emulate real-world clinical scenarios. The results demonstrated that the DCPN outperformed the prototypical network across all metrics, achieving the highest accuracies in same-domain tasks—70.86% for 1-shot, 82.57% for 5-shot, and 85.2% for 10-shot setups—corresponding to improvements of 5.51%, 5.72%, and 6.81%, respectively, over the prototypical network. Notably, in the same-domain 10-shot setting, the accuracy of the DCPN (85.2%) surpassed that of the PVT-based supervised learning model (85.15%), confirming its potential to diagnose rare diseases within few-shot learning frameworks.

1 July 2025 in Nature Communications
Verified
Journal Article
Open access
PDF file
TNTianhang Nan
Tianhang Nan
SZSong Zheng
Song Zheng
SQSiyuan Qiao
Siyuan Qiao
HQHao Quan
Hao Quan
XGXin Gao
Xin Gao
XCXiaoyu Cui
Xiaoyu Cui
Based on the expertise of pathologists, the pixelwise manual annotation has provided substantial support for training deep learning models of whole slide images (WSI)-assisted diagnostic. However, the collection of pixelwise annotation demands massive annotation time from pathologists, leading to a high burden of medical manpower resources, hindering to construct larger datasets and more precise diagnostic models. To obtain pathologists’ expertise with minimal pathologist workloads then achieve precise diagnostics, we collect the image review patterns of pathologists by eye-tracking devices. Simultaneously, we design a deep learning system: Pathology Expertise Acquisition Network (PEAN), based on the collected visual patterns, which can decode pathologists’ expertise and then diagnose WSIs. Eye-trackers reduce the time required for annotating WSIs to 4%, of the manual annotation. We evaluate PEAN on 5881 WSIs and 5 categories of skin lesions, achieving a high area under the curve of 0.992 and an accuracy of 96.3% on diagnostic prediction. This study fills the gap in existing models’ inability to learn from the diagnostic processes of pathologists. Its efficient data annotation and precise diagnostics provide assistance in both large-scale data collection and clinical care.

21 November 2025 in IEEE Transactions on Instrumentation and Measurement
Verified
Journal Article
Closed access
YDYong Ding
Yong Ding
LWLina Wang
Lina Wang
TNTianhang Nan
Tianhang Nan
YGYunqiu Gao
Yunqiu Gao
LLLisha Li
Lisha Li
XCXiaoyu Cui
Xiaoyu Cui
This paper proposes a high-precision human pose estimation method using sparse IMU sensors, featuring two novel components. Firstly, an anatomy-guided sensor association model is proposed. Based on the inherent anatomical relationships within the human body, the human body is divided into five main parts, with joints in each part being associated with specific IMUs. By means of this model, the limited data from sparse IMUs can be effectively utilized to enhance the accuracy of pose estimation. Secondly, a hybrid deep learning architecture is presented. It consists of five parallel branches corresponding to the five parts of the human body. The architecture combining Long Short-Term Memory (LSTM) and Kolmogorov-Arnold Networks (KAN) is utilized to reconstruct the pose of the corresponding anatomical part. The method addresses motion ambiguity caused by action distribution limitations in traditional inertial pose estimation. Experimental results demonstrate superior performance over state-of-the-art models across four public datasets.

19 March 2026 in IEEE/ASME Transactions on Mechatronics
Verified
Journal Article
Closed access
XCXiaoyu Cui
Xiaoyu Cui
YDYong Ding
Yong Ding
TNTianhang Nan
Tianhang Nan
YGYunqiu Gao
Yunqiu Gao
LLLisha Li
Lisha Li
YZYue Zhao
Yue Zhao
Existing glove-based finger motion tracking methods are often limited by site constraints, occlusion sensitivity, or insufficient feature representation. To address these issues, this article proposes an economical and efficient framework based on liquid metal sensor-embedded data gloves. Unlike prior glove-based approaches that rely on dense sensor arrays, the proposed framework uses a sparse array of 12 resistive liquid metal sensors (with rectangular and double-trapezoidal geometries tailored to hand anatomy) and fuses raw sensor signals with Fast Fourier Transform (FFT)-derived amplitude values to enhance the feature space, followed by a Transformer encoder for critical feature extraction and a feedforward neural network for motion mapping. Training data were collected using an optical hand motion tracking system (Leap Motion) as ground truth; ablation and comparative experiments demonstrate the framework achieves an average joint angle error of 1.50°, an R2 coefficient of 0.98, and an average processing latency of 0.03 ms, while operating without site constraints or occlusion concerns and maintaining stable performance across users with varying hand sizes (R2 ≥ 0.92), confirming its suitability for seamless human–computer interaction.