OBJECTIVE: This study introduces a novel evaluation framework, GPTRadScore, to systematically assess the performance of multimodal large language models (MLLMs) in generating clinically accurate findings from CT imaging. Specifically, GPTRadScore leverages LLMs as an evaluation metric, aiming to provide a more accurate and clinically informed assessment than traditional language-specific methods. Using this framework, we evaluate the capability of several MLLMs, including GPT-4 with Vision (GPT-4V), Gemini Pro Vision, LLaVA-Med, and RadFM, to interpret findings in CT scans. METHODS: This retrospective study leverages a subset of the public DeepLesion dataset to evaluate the performance of several multimodal LLMs in describing findings in CT slices. GPTRadScore was developed to assess the generated descriptions (location, body part, and type) using GPT-4, alongside traditional metrics. RadFM was fine-tuned using a subset of the DeepLesion dataset with additional labeled examples targeting complex findings. Post fine-tuning, performance was reassessed using GPTRadScore to measure accuracy improvements. RESULTS: Evaluations demonstrated a high correlation of GPTRadScore with clinician assessments, with Pearson's correlation coefficients of 0.87, 0.91, 0.75, 0.90, and 0.89. These results highlight its superiority over traditional metrics, such as BLEU, METEOR, and ROUGE, and indicate that GPTRadScore can serve as a reliable evaluation metric. Using GPTRadScore, it was observed that while GPT-4V and Gemini Pro Vision outperformed other models, significant areas for improvement remain, primarily due to limitations in the datasets used for training. Fine-tuning RadFM resulted in substantial accuracy gains: location accuracy increased from 3.41% to 12.8%, body part accuracy improved from 29.12% to 53%, and type accuracy rose from 9.24% to 30%. These findings reinforce the hypothesis that fine-tuning RadFM can significantly enhance its performance. CONCLUSION: GPT-4 effectively correlates with expert assessments, validating its use as a reliable metric for evaluating multimodal LLMs in radiological diagnostics. Additionally, the results underscore the efficacy of fine-tuning approaches in improving the descriptive accuracy of LLM-generated medical imaging findings.

In chest X-ray (CXR) image analysis, rule-based systems are usually employed to extract labels from reports for dataset releases. However, there is still room for improvement in label quality. These labelers typically output only presence labels, sometimes with binary uncertainty indicators, which limits their usefulness. Supervised deep learning models have also been developed for report labeling but lack adaptability, similar to rule-based systems. In this work, we present MAPLEZ (Medical report Annotations with Privacy-preserving Large language model using Expeditious Zero shot answers), a novel approach leveraging a locally executable Large Language Model (LLM) to extract and enhance findings labels on CXR reports. MAPLEZ extracts not only binary labels indicating the presence or absence of a finding but also the location, severity, and radiologists' uncertainty about the finding. Over eight abnormalities from five test sets, we show that our method can extract these annotations with an increase of 3.6 percentage points (pp) in macro F1 score for categorical presence annotations and more than 20 pp increase in F1 score for the location annotations over competing labelers. Additionally, using the combination of improved annotations and multi-type annotations in classification supervision, we demonstrate substantial advancements in model quality, with an increase of 1.1 pp in AUROC over models trained with annotations from the best alternative approach. We share code and annotations.

We consider the problem of estimating high-dimensional and nonparametric distributions in distributed networks, where each sensor in the network observes an independent sample from the underlying distribution and can communicate it to a central processor by writing at most k bits on a public blackboard. We obtain matching upper and lower bounds for the minimax risk of estimating the underlying distribution under L1loss. Our results reveal that the minimax risk reduces exponentially in k. Instead of relying on strong data processing inequalities for the converse as commonly done in the literature, we build on a new representation of the communication constraint, which leads to a tight characterization of the problem.

We consider a two-user multiple-input multiple-output multiple access wiretap channel with N antennas at each transmitter, N antennas at the legitimate receiver, and K antennas at the eavesdropper. We determine the optimal sum secure degrees of freedom (s.d.o.f.) for this model for all values of N and K. We subdivide our problem into several regimes based on the values of N and K, and provide achievable schemes based on vector space alignment and real alignment techniques for fixed and fading channel gains. To prove the optimality of the achievable schemes, we provide matching converses for each regime. Our results show how the number of eavesdropper antennas affects the optimal sum s.d.o.f. of the multiple access wiretap channel.

We consider two fundamental multi-user channel models: the multiple-input multiple-output (MIMO) wiretap channel with one helper (WTH) and the MIMO multiple access wiretap (MAC-WT) channel. In each case, the eavesdropper has K antennas while the remaining terminals have N antennas each. We consider a fast fading channel where the channel state information (CSI) of the legitimate receiver is available at the transmitters but no CSI at the transmitters (CSIT) is available for the eavesdropper's channel. We determine the optimal sum secure degrees of freedom (s.d.o.f.) for each channel model for the regime K ≤ N, and show that in this regime, the MAC-WT channel reduces to the WTH in the absence of eavesdropper CSIT. For the regime N ≤ K ≤ 2N, we obtain the optimal linear s.d.o.f., and show that the MAC-WT channel and the WTH have the same optimal s.d.o.f. when restricted to linear encoding strategies. In the absence of any such restrictions, we provide an upper bound for the sum s.d.o.f. of the MAC-WT channel in the regime N ≤ K ≤ 2N. Our results show that unlike in the single-input single-output case, there is loss of s.d.o.f. for even the WTH due to lack of eavesdropper CSIT when K ≥ N.

The two user multiple-input single-output (MISO) broadcast channel with confidential messages (BCCM) is studied, in which the nature of channel state information at the transmitter (CSIT) from each user can be of the form Ii, i = 1, 2 where I1, I2∈ {P, D, N}, and the forms P, D, and N correspond to perfect and instantaneous, completely delayed, and no CSIT, respectively. Thus, the overall CSIT can alternate between nine possible states corresponding to all possible values of I1I2, with each state occurring for λI1I2fraction of the total duration. We assume that perfect and instantaneous CSI is available at the all receivers. The main contribution of this paper is to establish the secure degrees of freedom (s.d.o.f.) region of the MISO BCCM with alternating CSIT with the symmetry assumption, where λI1I2= λI2I1. The main technical contributions include developing 1) novel achievable schemes for MISO BCCM with alternating CSIT with security constraints, which also highlight the synergistic benefits of inter-state coding for secrecy; 2) new converse proofs via local statistical equivalence and channel enhancement; and 3) showing the interplay between various aspects of channel knowledge and their impact on s.d.o.f.

We consider three channel models: the wiretap channel with M helpers, the K-user multiple access wiretap channel, and the K-user interference channel with an external eavesdropper, when no eavesdropper's channel state information (CSI) is available at the transmitters. In each case, we establish the optimal sum secure degrees of freedom (s.d.o.f.) by providing achievable schemes and matching converses. We show that the unavailability of the eavesdropper's channel state information at the transmitter (CSIT) does not reduce the s.d.o.f. of the wiretap channel with helpers. However, there is loss in s.d.o.f. for both the multiple access wiretap channel and the interference channel with an external eavesdropper. In particular, we show that in the absence of eavesdropper's CSIT, the K-user multiple access wiretap channel reduces to a wiretap channel with (K - 1) helpers from a sum s.d.o.f. perspective, and the optimal sum s.d.o.f. reduces from K(K-1) /K(K-1)+1 to K-1 K . For the interference channel with an external eavesdropper, the optimal sum s.d.o.f. decreases from K(K-1) /2K-1 to K-1/ 2 in the absence of the eavesdropper's CSIT. Our results show that the lack of eavesdropper's CSIT does not have a significant impact on the optimal s.d.o.f. for any of the three channel models, especially when the number of users is large. This implies that physical layer security can be made robust to the unavailability of eavesdropper CSIT at high signal-to-noise ratio regimes by the careful modification of the achievable schemes as demonstrated in this paper.

We consider a two-user multiple-input multiple-output (MIMO) multiple access wiretap channel with N antennas at each transmitter, N antennas at the legitimate receiver, and K antennas at the eavesdropper. We determine the optimal sum secure degrees of freedom (s.d.o.f.) when the channel gains are fixed for the duration of the communication. We provide optimal achievable schemes based on a combination of Gaussian signaling and real interference alignment for all regimes of N and K.

We present a real interference alignment technique for multiple-input multiple-output (MIMO) networks. This technique is based on a theorem due to Dirichlet and Khintchine for simultaneous Diophantine approximation and uses the outputs of all the antennas at the receiver simultaneously for decoding, instead of using them in an antenna-by-antenna basis. This allows us to forgo asymptotic real interference alignment for several multi-user scenarios such as the two-user MIMO interference channel with confidential messages and the two-user MIMO multiple access wiretap channel.

The two-user multiple-input single-output (MISO) broadcast channel with confidential messages (BCCM) is studied in which the nature of channel state information at the transmitter (CSIT) from each user can be of the form Ii, i = 1, 2 where I1; I2∈ {P;D;N}, and the forms P, D and N correspond to perfect and instantaneous, completely delayed, and no CSIT, respectively. Thus, the overall CSIT can alternate over time between 9 possible states corresponding to all possible values of I1I2, with each state occurring for λI1I2fraction of the total duration. The main contribution of this paper is to establish the secure degrees of freedom (s.d.o.f.) region of the MISO BCCM with alternating CSIT with the symmetry assumption λI1I2= λI2I1. The results highlight the synergistic benefits of coding across CSIT states for secrecy and the interplay between various aspects of channel knowledge and its impact on s.d.o.f.

We consider the two-user multiple-input single-output (MISO) broadcast channel with confidential messages (BCCM), in which the nature of channel state information at the transmitter (CSIT) from each user can be of the form P, D and N, corresponding to perfect and instantaneous, completely delayed, and no CSIT, respectively. We focus on the cases with heterogeneous CSIT from the users, that is, the states PD, PN and DN. The main contribution of this paper is to establish the exact secure degrees of freedom (s.d.o.f.) regions of the MISO BCCM in all of these three heterogeneous states. The results highlight the impact of availability of CSIT on the s.d.o.f. region.

We consider the K-user Gaussian multiple access wiretap channel (MAC-WT), where no eavesdropper channel state information (CSI) is available at the transmitters. We show that the exact sum secure degrees of freedom (s.d.o.f.) of this channel model is K-1/K . This result shows that, under the condition of no eavesdropper CSI, the MAC-WT acts like a single-transmitter K - 1 helper wiretap channel. We further show that, when a subset of the transmitters have eavesdropper CSI, then higher sum s.d.o.f. can be achieved, and the system can be operated as a MAC-WT for the users with eavesdropper CSI, with the remaining users acting as helpers. In particular, if m of the K transmitters have eavesdropper CSI, we show that m(K-1)/m(K-1)+1 sum s.d.o.f. can be achieved, showing the benefits of having the eavesdropper CSI at the transmitters.

We consider the block Rayleigh fading multiple-input multiple-output (MIMO) wiretap channel with no prior channel state information (CSI) available at any of the terminals. The channel gains remain constant within a coherence interval of T symbols, and then change to another independent realization in the next coherence interval. The transmitter, the legitimate receiver, and the eavesdropper have nt, nr, and ne antennas, respectively. We determine the exact secure degrees of freedom (s.d.o.f.) of this system when T ≥ 2min(nt,nr). We show that, in this case, the s.d.o.f. is exactly equal to (min(nt,nr)-ne)+(T -min(nt,nr))/T. The first term in this expression can be interpreted as the eavesdropper with ne antennas taking away ne antennas from both the transmitter and the legitimate receiver. The second term can be interpreted as a fraction of the s.d.o.f. being lost due to the lack of CSI at the legitimate receiver. In particular, the fraction loss, min(nt,nr)/T, can be interpreted as the fraction of channel uses dedicated to training the legitimate receiver for it to learn its own CSI. We prove that this s.d.o.f. can be achieved by employing a constant norm channel input, which can be viewed as a generalization of discrete signalling to multiple dimensions.

We consider the block Rayleigh fading multiple-input multiple-output (MIMO) wiretap channel with no prior channel state information (CSI) available at any of the terminals. The channel gains remain constant in a coherence time of T symbols, and then change to another independent realization. The transmitter, the legitimate receiver and the eavesdropper have nt, nrand neantennas, respectively. We determine the exact secure degrees of freedom (s.d.o.f.) of this system when T ≥ 2 min(nt, nr). We show that, in this case, the s.d.o.f. is exactly (min(nt, nr) − ne)+(T − min(nt, nr))/T. The first term can be interpreted as the eavesdropper with neantennas taking away neantennas from both the transmitter and the legitimate receiver. The second term can be interpreted as a fraction of s.d.o.f. being lost due to the lack of CSI at the legitimate receiver. In particular, the fraction loss, min(nt, nr)/T, can be interpreted as the fraction of channel uses dedicated to training the legitimate receiver for it to learn its own CSI. We prove that this s.d.o.f. can be achieved by employing a constant norm channel input, which can be viewed as a generalization of discrete signalling to multiple dimensions.

We study the two-user multiple-input single-output (MISO) broadcast channel with confidential messages under the assumption of alternating channel state information at the transmitter (CSIT). We consider two alternating states: PD and DP which occur for an equal fraction of time. In state PD, the CSIT of the channel to the first receiver is available perfectly without delay (P) while that of the second receiver is available with a delay of one channel use (D); in state DP, the roles of the receivers are reversed. We characterize the exact secure degrees of freedom (s.d.o.f.) region of this system, and show as a corollary that the sum s.d.o.f. is 3/2. We observe that this sum s.d.o.f. is the same as what can be achieved by the states PP and DD occurring for equal fraction of time. Though the s.d.o.f. of the system in the states PD and DP is not known individually, we are able to establish the s.d.o.f. region when the two states alternate and occur for an equal fraction of the time.

We consider the fast Rayleigh fading wiretap channel, over which a legitimate transmitter wishes to have secure communication with a legitimate receiver in the presence of an eavesdropper. We consider an average power constraint on the input, and assume that no channel state information (CSI) is available to any user. We show that the input distribution that achieves the secrecy capacity for this wiretap channel is discrete with a finite number of mass points.

This paper presents a bit serial realization of the sign-LMS based adaptive filter which enjoys multiplier free weight update loop. To reduce the complexity of the multipliers that arise in the filtering process, the filter weights are represented in the so-called canonic SPT form which guarantees presence of at least one zero between every two non-zero power-of-two terms. As the filter weights are not fixed but updated in time, it is essential to ensure that the canonic SPT format is retained in the updated filter coefficients. For this, a bit serial adder is proposed that takes as input two numbers in canonic SPT and produces an output also in canonic SPT. It is further shown how the canonic SPT property of the input can be used to reduce the complexity of the adder. For the filtering part, a bit serial multiplier is developed that takes one input (i.e., data bits) in 2's complement form and the other input (i.e., weight bits) in canonic SPT, producing the result in 2's complement. The multiplication can not, however, be realized using a few fixed shift and add operations, since the position of the non-zero SPT terms in the canonic SPT expression of each coefficient changes with time. The proposed multiplier instead multiplies the 2's complement number with pairs of consecutive SPT bits of the other number. The resulting partial products can be realized using simple AND-OR logic.