Automated analysis of temporal changes in multimodal retinal images is critical for the prognostic assessment of ophthalmic diseases. Unlike traditional single-timepoint diagnosis, tracking longitudinal changes across multiple imaging modalities introduces significant data bias challenges: (1) Imbalanced modality samples compromise the integration of knowledge within minority modalities; (2) Heterogeneous visual patterns across modalities undermine the perception of disease-relevant biomarkers. To tackle these issues, we propose a Modality-Incremental Expert Aggregation Network (MoEA-Net), which unifies the inter-modal integration and intra-modal perception for enhanced retinal prognostic prediction. Specifically, we employ the large language model (LLM) with incremental LoRA layers for specific modalities to effectively integrate knowledge from imbalanced data. Besides, we introduce a Spatiotemporal-aware Expert (SAE) module to better perceive both the anatomical structures and longitudinal changes within modalities. By progressively combining the SAE module with incremental LoRA, MoEA-Net supports continual knowledge accumulation and improves accurate reasoning. Experimental results show that MoEA-Net achieves state-of-the-art performance on subretinal fluid change and visual recovery classification tasks, validating its effectiveness.

The automatic generation of brain CT reports has gained widespread attention, given its potential to assist radiologists in diagnosing cranial diseases. However, brain CT scans involve extensive medical entities, such as diverse anatomy regions and lesions, exhibiting highly inconsistent spatial patterns in 3D volumetric space. This leads to biased learning of medical entities in existing methods, resulting in repetitiveness and inaccuracy in generated reports. To this end, we propose a Medical Entity-balanced Prompting Network (MEPNet), which harnesses the large language model (LLM) to fairly interpret various entities for accurate brain CT report generation. By introducing the visual embedding and the learning status of medical entities as enriched clues, our method prompts the LLM to balance the learning of diverse entities, thereby enhancing reports with comprehensive findings. First, to extract visual embedding of entities, we propose Knowledge-driven Joint Attention to explore and distill entity patterns using both explicit and implicit medical knowledge. Then, a Learning Status Scorer is designed to evaluate the learning of entity visual embeddings, resulting in unique learning status for individual entities. Finally, these entity visual embeddings and status are elaborately integrated into multi-modal prompts, to guide the text generation of LLM. This process allows LLM to self-adapt the learning process for biased-fitted entities, thereby covering detailed findings in generated reports. We conduct experiments on two brain CT report generation benchmarks, showing the effectiveness in clinical accuracy and text coherence.

Multichannel parameter estimation is a key technology for azimuth multichannel terrain observation by progressive scans synthetic aperture radar (MC TOPS SAR) systems. Normally, for the traditional multichannel parameter estimation algorithms, the mismatch parameters are concluded as phase error, baseline error, and Doppler centroid. The beam rotation speed error is ignored. However, for MC TOPS SAR, beam rotation speed error will degrade the reconstruction performance of the Doppler spectrum. Moreover, only part of the mismatch parameters can be estimated for the traditional algorithms. To address those problems, a joint parameters estimation algorithm for TOPS SAR based on spatial time cross correlation coefficient (STCCC) is proposed. Combined with phase wrapping and the least-squares method (LSM), the algorithm is capable of realizing a joint parameters estimation of channel consistency error, azimuth baseline error, Doppler centroid frequency, and beam rotation speed estimation without iteration. The real data processing result shows that the method can effectively realize MC TOPS SAR data error estimation.

Synthetic aperture radar (SAR) sometimes needs to be imaged with small time bandwidth product (TBP) for fast observation. As a real-time image algorithm, back-projection algorithm (BPA) is efficient for fast observation. However, compared with the frequency-domain algorithm, the influence of small TBP (STBP) on BPA has not been fully discussed. Thus, the performance of the original BPA on STBP SAR imaging is analyzed and an improved BPA for STBP imaging is proposed in this letter. Theoretical analysis indicates that STBP will cause the point spread function (PSF) distortion and azimuth spectrum ambiguity to the BPA imaging result. To deal with those problems, two parameters of the original BPA are modified in the improved BPA. The accumulating azimuth angle (AAA) is optimized first to calibrate the PSF distortion. Then, based on the azimuth-ambiguity-to-signal ratio (AASR), a BPA meshing criterion related to TBP is proposed to diminish azimuth spectrum ambiguity. The simulation and real data results demonstrate that the improved BPA is effective on undistorted PSF recovering and unambiguity spectrum maintaining, even when the TBP is down to 4.

Combined with terrain observation by progressive scans (TOPS) synthetic aperture radar (SAR), near-nadir SAR (N-SAR) system has great potential for the surface water and ocean topography (SWOT) observation. However, one practical problem of the airborne N-SAR is the non-ideal attitude angle caused by the environment. Normally, the influence of attitude angle error can be considered as a constant squint angle of the system. However, in fact, for the near-nadir SAR system with TOPS SAR mode, the attitude angle error will cause a non-ideal 2-dimensional space-variant Doppler centroid for the receiving data and cause the traditional imaging algorithm failure. To deal with such problems, an imaging algorithm with attitude angle estimation for the near-nadir TOPS SAR system is proposed in this paper. Firstly, the influence of attitude angle error and the signal properties are analyzed. Secondly, based on the analysis, an attitude angle estimate algorithm (AAE), utilizing the nonlinear least square method (NLSM) and minimum entropy criterion, is proposed. Then, based on the estimated altitude angle, without data blocking, a full data imaging algorithm (FDA) is proposed. Its main idea is to obtain an un-ambiguous azimuth spectrum and design an appropriate range-variant azimuth filter through frequency chirp scaling (FCS). The numerical simulations and real data processing demonstrate that the proposed algorithm has good performance on attitude angle error estimation and well-focused image obtaining.

Combined with Terrain Observation by Progressive Scans (TOPS) Synthetic Aperture Radar (SAR), Near-nadir Interferometry SAR (NInSAR) has great potential on the Surface Water and Ocean Topography (SWOT) observation. One practical problem is that the Doppler center of airborne NInSAR varies along range and its bandwidth is greater than pulse repeat frequency (PRF) due to the affection of yaw angle error. Traditional imaging algorithms divide the data into several range blocks so that the range variation of Doppler center is ignorable. However, the drawback is that the size and the overlap of the blocks are hard to design. In this paper, an improved image algorithm without range block is proposed. The main procedure is to obtain unambiguous azimuth spectrum and design appropriate range varied azimuth filter through frequency chirp scaling algorithm. The good performance of the proposed algorithm is demonstrated through point target simulation and real data.