Few-Shot Classification and Anatomical Localization of Tissues in SPECT Imaging
Mohammed Abdul Hafeez Khan, Samuel Morries Boddepalli, Siddhartha Bhattacharyya, Debasis Mitra
TL;DR
The paper tackles tissue classification and anatomical localization in SPECT when labeled data are scarce by adapting two neural approaches: Prototypical Networks for few-shot classification and PRNet for 2D landmark localization. Prototypical Networks map inputs to a metric space and classify by class prototypes, using a distance-based probability $p_\phi(y = k | x)$ and prototypes $c_k$ computed from support features. PRNet is reformatted to 2D SPECT with an encoder-decoder and skip connections, trained with a self-supervised loss $L_{ssl} = L_{dis} + L_{rec}$ to learn spatial relationships between patches. Results demonstrate high classification accuracy on a heart-centered 2D SPECT slice and meaningful spatial localization signals, suggesting potential to enhance segmentation and localization with limited data in clinical imaging.
Abstract
Accurate classification and anatomical localization are essential for effective medical diagnostics and research, which may be efficiently performed using deep learning techniques. However, availability of limited labeled data poses a significant challenge. To address this, we adapted Prototypical Networks and the Propagation-Reconstruction Network (PRNet) for few-shot classification and localization, respectively, in Single Photon Emission Computed Tomography (SPECT) images. For the proof of concept we used a 2D-sliced image cropped around heart. The Prototypical Network, with a pre-trained ResNet-18 backbone, classified ventricles, myocardium, and liver tissues with 96.67% training and 93.33% validation accuracy. PRNet, adapted for 2D imaging with an encoder-decoder architecture and skip connections, achieved a training loss of 1.395, accurately reconstructing patches and capturing spatial relationships. These results highlight the potential of Prototypical Networks for tissue classification with limited labeled data and PRNet for anatomical landmark localization, paving the way for improved performance in deep learning frameworks.