Transferring Knowledge from High-Quality to Low-Quality MRI for Adult Glioma Diagnosis

TL;DR

The approach provides insights into improving glioma diagnosis in SSA, showing the potential of deep learning in resource-limited settings and the importance of transfer learning from high-quality datasets.

Abstract

Glioma, a common and deadly brain tumor, requires early diagnosis for improved prognosis. However, low-quality Magnetic Resonance Imaging (MRI) technology in Sub-Saharan Africa (SSA) hinders accurate diagnosis. This paper presents our work in the BraTS Challenge on SSA Adult Glioma. We adopt the model from the BraTS-GLI 2021 winning solution and utilize it with three training strategies: (1) initially training on the BraTS-GLI 2021 dataset with fine-tuning on the BraTS-Africa dataset, (2) training solely on the BraTS-Africa dataset, and (3) training solely on the BraTS-Africa dataset with 2x super-resolution enhancement. Results show that initial training on the BraTS-GLI 2021 dataset followed by fine-tuning on the BraTS-Africa dataset has yielded the best results. This suggests the importance of high-quality datasets in providing prior knowledge during training. Our top-performing model achieves Dice scores of 0.882, 0.840, and 0.926, and Hausdorff Distance (95%) scores of 15.324, 37.518, and 13.971 for enhancing tumor, tumor core, and whole tumor, respectively, in the validation phase. In the final phase of the competition, our approach successfully secured second place overall, reflecting the strength and effectiveness of our model and training strategies. Our approach provides insights into improving glioma diagnosis in SSA, showing the potential of deep learning in resource-limited settings and the importance of transfer learning from high-quality datasets.

Figures (5)

• Figure 1: Representative slices respectively in axial, coronal, and sagittal views from two datasets. The upper slices are from the high-quality RSNA-ASNR-MICCAI BraTS-GLI 2021 dataset, and the bottom slices are from the low-quality BraTS-Africa dataset.
• Figure 2: The process of model training and segmentation inference. During the model training process, three data usage strategies are employed to separately train the original network and the larger network. In the inference phase, the outputs of the two networks are ensembled to obtain the final segmentation.
• Figure 3: Sagittal slices of the same subject before and after applying super-resolution techniques. The data after enhancement with super-resolution techniques appears smoother, with a more pronounced contrast of internal brain structures.
• Figure 4: Comparison of segmentation results using the three strategies—$S_{GLI\rightarrow SSA}$, $S_{SSA}$, and $S_{srSSA}$ on selected validation samples. The color coding indicates the non-enhancing tumor core (red), surrounding non-enhancing FLAIR hyperintensity (green), and enhancing tumor (blue).
• Figure 5: Segmentation results from final submission on selected examples. (a) shows the results on the training data, and (b) on the validation data. The color coding indicates the non-enhancing tumor core (red), surrounding non-enhancing FLAIR hyperintensity (green), and enhancing tumor (blue).