| 講演抄録/キーワード |
| 講演名 |
2025-12-05 14:00
中皮腫細胞診断のための白色光散乱スペクトルと光学顕微鏡画像に基づいたマルチモーダル手法の開発 ○羅 向昆(奈良先端大)・崇風まあぜん(奈良先端大/宮崎大)・釣 優香・谷 懿(奈良先端大)・武内風香・伊藤彰彦(近畿大)・細川陽一郎・大竹義人(奈良先端大) MICT2025-45 MI2025-53 |
| 抄録 |
(和) |
The cytological diagnosis of malignant mesothelioma is challenging due to its morphological overlap with the reactive mesothelial cells. This study proposes a multimodal deep learning approach integrating white-light scattering spectrum and cellular morphology derived from optical microscope images for the diagnosis of malignant mesothelioma. We analyzed image and spectral data of 261 cells from 10 patients (five reactive mesothelial cells, and five malignant mesothelioma). Three models were attempted: a spectrum-only model (PCA+RF), an image-only model (fine-tuned ResNet50), and a multimodal model using a gated fusion mechanism. The models were validated in a leave-two-patients-out cross-validation. While the Image-only model achieved the highest AUC (0.970) and Precision (0.941), it suffered from low Recall (0.799). Our proposed multimodal approach, however, achieved the best overall Accuracy (0.907±0.121) and F1-Score (0.899±0.126). It demonstrated a clear synergistic effect by creating a more balanced classifier, dramatically improving Recall to 0.949±0.068. These findings highlight the potential of data fusion strategies to create more accurate and reliable tools for cytological diagnosis. |
| (英) |
The cytological diagnosis of malignant mesothelioma is challenging due to its morphological overlap with the reactive mesothelial cells. This study proposes a multimodal deep learning approach integrating white-light scattering spectrum and cellular morphology derived from optical microscope images for the diagnosis of malignant mesothelioma. We analyzed image and spectral data of 261 cells from 10 patients (five reactive mesothelial cells, and five malignant mesothelioma). Three models were attempted: a spectrum-only model (PCA+RF), an image-only model (fine-tuned ResNet50), and a multimodal model using a gated fusion mechanism. The models were validated in a leave-two-patients-out cross-validation. While the Image-only model achieved the highest AUC (0.970) and Precision (0.941), it suffered from low Recall (0.799). Our proposed multimodal approach, however, achieved the best overall Accuracy (0.907±0.121) and F1-Score (0.899±0.126). It demonstrated a clear synergistic effect by creating a more balanced classifier, dramatically improving Recall to 0.949±0.068. These findings highlight the potential of data fusion strategies to create more accurate and reliable tools for cytological diagnosis. |
| キーワード |
(和) |
Malignant Mesothelioma / Multimodal / Deep Learning / White-light Scattering / Cytological Diagnosis / / / |
| (英) |
Malignant Mesothelioma / Multimodal / Deep Learning / White-light Scattering / Cytological Diagnosis / / / |
| 文献情報 |
信学技報, vol. 125, no. 270, MI2025-53, pp. 54-57, 2025年12月. |
| 資料番号 |
MI2025-53 |
| 発行日 |
2025-11-27 (MICT, MI) |
| ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
| PDFダウンロード |
MICT2025-45 MI2025-53 |
|