Abstract 1301: Multiple-channel hybrid neural network to achieve quantitative bioluminescence tomography for cancer research

Abstract Purpose: Bioluminescence tomography(BLT) enhances commonly used 2D bioluminescence imaging by reconstructing 3D distributions of bioluminescence activity, allowing in vivo tumor localization. However, traditional physics model-based BLT struggles with model uncertainties and ill-posedness, limiting its ability to provide quantitative information, such as tumor volume. To address this, we introduce the Multiple-channel Hybrid Neural Network(McHyNN), a machine learning(ML) framework integrating light propagation model, multimodal imaging, and cross-animal data. We demonstrate McHyNN’s ability to accurately identify tumors and quantify their volumes, particularly in cases of small and multiple brain metastases(BM), making it highly valuable for cancer research. Method: McHyNN is a feed-forward neural network(NN) grounded in a light propagation model based on the diffusion equation. It incorporates multi-data channels, including multi-animal and multimodal data. For a given animal set, the data includes BL images emitted from internal tumors and corresponding MRI scan, if tumor is MRI-detectable. The NN outputs n+1 BL tumor distributions, including the n existing animal sets and the new subject, whose tumor information is to be determined. McHyNN leverages mutual information from light propagation across multiple animals and multimodal images, and the NN is optimized accordingly using loss function L = Ldiff + Limage. Ldiff and Limage minimize the difference between calculated and actual BL tumor distributions based on surface BLI measurement, and MRI data, respectively. We will first present numerical results based on simulated brain metastases in mice with experimentally determined noise, followed by showing the capability of BLT to reconstruct in vivo BMs in SCLC spontaneous animal model. Results: McHyNN accurately quantifies three small BMs (1.2 to 1.9 mm in length, volume: 0.9 to 1.9 mm3) within simulated mouse brain at localization accuracy of < 0.5mm and volume errors within 0.8 mm3, even when the tumors are separated by only 2-2.7 mm. In contrast, the compressive sensing algorithm commonly used in conventional BLT fails to distinguish these tumors. Moreover, we simulated single brain metastasis ranging from 1.0-3.6 mm in length, 0.4-17.2 mm3 including 10 tumors 1.5<mm or="" 1.75="" mm3. McHyNN successfully localized all cases within 0.5 mm and faithfully reconstructed tumor volumes, showing a linear correlation with ground truth (R2=0.99). The in vivo BMs study is ongoing. <mm or="" 1.75="" mm <mm or="" 1.75="" mm. Conclusions: Our BLT platform incorporating the McHyNN approach has potential to transform qualitative bioluminescence imaging into a 3D quantitative modality, particularly for tumor volume quantification and detecting small malignancies. This innovation will significantly enhance investigators’ capabilities in cancer studies. Citation Format: Beichuan Deng, Zhishen Tong, Yi-chun Tsai, James Kim, Hamid Dehghani, Ken Kang-Hsin Wang. Multiple-channel hybrid neural network to achieve quantitative bioluminescence tomography for cancer research [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 1301.