On July 1st 2024 at 16:00, Panagiotis Gonidakis will defend their PhD entitled “DATA- AND LABEL-EFFICIENT DEEP LEARNING FOR MEDICAL IMAGE ANALYSIS APPLICATION TO LUNG NODULE DETECTION ON THORACIC CT”.

Everybody is invited to attend the presentation in room D.0.03, or digitally via this link.

Convolutional neural networks (CNNs) have been widely used to detect and classify various objects and structures in computer vision and medical imaging. Access to large sets of annotated data is commonly a prerequisite for achieving good performance. In medical imaging, acquiring adequate amounts of labelled data can often be time consuming and costly. Therefore, reducing the need for data and in particular associated annotations, is of high importance for medical imaging applications. In this work we investigated whether we can lower the need of annotated data for a supervised learning classification problem.  

We chose to tackle the problem of lung nodule detection in thoracic computed tomography (CT) imaging, as this widely investigated application allowed us to benefit from publicly available data and benchmark our methods. We designed a 3D CNN architecture to perform patch-wise classification of candidate nodules for false positive reduction. Its training, testing and fine-tuning procedure is optimized, we evaluated its performance, and we compared it with other state-of-the-art approaches in the field.  

Next, we explored how data augmentation can contribute towards more accurate and less data-demanding models. We investigated the relative benefit of increasing the amount of original data, with respect to computationally augmenting the amount of training samples. Our result indicated that in general, better performance is achieved when increasing the amount of unique data samples, or augmenting the data more extensively, as expected. Surprisingly however, we observed that after reaching a certain amount of training samples, data augmentation led to significantly better performance compared to adding unique samples. Amongst investigated augmentation methods, rotations were found to be most beneficial for improving model performance. 

Following, we investigated the benefit of combining deep learning with handcrafted features. We explored three fusion strategies with increasing complexity and assessed their performance for varying amounts of training data. Our findings indicated that combining handcrafted features with a 3D CNN approach significantly improved lung nodule detection performance in comparison to an independently trained CNN model, regardless of the fusion strategy. Comparatively larger increases in performance were obtained when less training data was available. The fusion strategy in which features are combined with a CNN using a single end-to-end training scheme performed best overall, allowing to reduce training data by 33% to 43%, while maintaining performance. Among the investigated handcrafted features, those that describe the relative position of the candidate with respect to the lung wall and mediastinum, were found to be of most benefit. 

Finally, we considered the case in which abundant data is available, but annotations are scarce, and investigated several methods to improve label-efficiency and their combined effect. We proposed a framework that utilizes both annotated and unannotated data, can be pretrained via self-supervision, and allows to combine handcrafted features with learned representations. Interestingly, the improvements in performance derived from the proposed learning schemes were found to accumulate, leading to increased label-efficiency when these strategies are combined. We observed a potential to decrease the amount of annotated data up to 68% when compared to traditional supervised training, while maintaining performance.  

Our findings indicate that the investigated methods allow considerable reduction of data and/or annotations while maintaining model performance for lung nodule detection from CT imaging. Future work should investigate whether these results generalize to other domains, such that more applications that face challenges due to a shortage of annotated data may benefit from the potential of deep learning.

Zamzam won the award for the third place in the Best Student Paper category at the World Congress on Medical Informatics in Sydney for the paper “Baseline Survey on Referrals and Healthcare Provider Needs in View for an Electronic Referral System” authored by Zamzam Kalume, Bart Jansen, Marc Nyssen, Jan Cornelis and Frank Verbeke .

On November 15th 2024 at 10:00, Eden Teshome Hunde will defend their PhD entitled “CROSS-LAYER DESIGN, IMPLEMENTATION AND EVALUATION OF IPV6 MULTICAST FOR RADIO DUTY CYCLED WIRELESS SENSOR AND ACTUATOR NETWORKS”.

Everybody is invited to attend the presentation in room D.2.01 or online via this link.

In this work, we study Bidirectional Multicast RPL Forwarding (BMRF) as this protocol relies on forwarding tables put in place by the well-known Routing Protocol for Low Power and Lossy Networks (RPL) and allows to combine the best ideas of existing multicast protocols. Through RPL, a routing tree towards the sink is installed for multihop routing from node to sink, and the nodes’ forwarding tables will also contain entries for reaching destinations in downward direction.

For downward forwarding IPv6 multicast packets, two methods exist. One is via link layer (LL), broadcasting a frame containing the IPv6 multicast packet. The other is to send several LL unicast frames containing that packet. BMRF allows a node to choose between these two methods. The best option will depend on the presence of a radio duty cycling (RDC) protocol. RDC is part of the medium access control (MAC) layer and puts the radio to sleep when no communication is needed. We investigate the influence of MAC/RDC protocols on BMRF’s performance.

We evaluate the performance of BMRF on non-synchronized WSANs that use Carrier Sense Multiple Access (CSMA) as MAC and ContikiMAC as RDC. We demonstrate that LL unicast outperforms LL broadcast in terms of packet delivery ratio (PDR), delay, and energy consumption in many settings.

We investigate the performance of BMRF on WSANs with synchronous MAC and RDC based on Time Slotted Channel hopping (TSCH). This is more challenging, as TSCH needs a schedule to tell which action must happen in each timeslot. The actions can be to send or to listen on a given channel or to be idle. Idleness allows the radio to switch OFF, providing RDC. The schedule is not part of the standard and must be proposed by the system designer. An elegant autonomous scheduling method called Orchestra is available to accommodate traffic in a RPL tree. We extend Orchestra with a novel scheduling rule for supporting LL downwards forwarding through LL broadcast. Comparing LL unicast with LL broadcast forwarding teaches us that LL unicast outperforms LL broadcast in terms of packet delivery ratio (PDR), but the latter can be beneficial to certain applications, especially those sensitive to delay.

Before conducting the two previous evaluation studies, we investigate the performance of simple convergecast traffic while considering ContikiMAC and TSCH with Orchestra under RPL on the real dual Zolertia Firefly Motes (one is observed and other one is observing mote). This study served two purposes; it reminds the reader of the characteristics of those protocols and allowed to fine-tune the dual motes.

We also contributed by adapting the Orchestra to bursty convergecast traffic. Simulation results demonstrate that the new scheduler slightly improves PDR and reduces delay compared to state-of-the-art solutions.

On May 16th 2025 at 16:00, Pooria Iranian will defend their PhD entitled “FLUORESCENCE-LIFETIME ENDOSCOPY WITH A TIME-GATED CAMERA”.

Everybody is invited to attend the presentation in room I.0.01 or online via this link.

The diagnosis and treatment of cancer remain complex and challenging, particularly during surgical procedures where accurately distinguishing between malignant and healthy tissue is critical. Conventional imaging modalities such as MRI and CT often fail to provide the real-time, high-contrast visualization necessary for precise tumour removal.

Fluorescence imaging has emerged as a valuable technique, offering real-time, high-sensitivity visualization of tissues. However, traditional fluorescence-intensity imaging is limited by the fluorophore’s intensity and differentiating between fluorophores with overlapping emission spectra. In contrast, Fluorescence Lifetime Imaging (FLT) provides a more reliable alternative. FLT is less susceptible to variations in fluorophore concentration and offers a more accurate representation of the tissue environment.

To enhance the accuracy of tumour resection during image-guided surgeries, our research team is developing an innovative imaging system based on the Current-Assisted Photonic Sampler (CAPS). This advanced camera enables real-time FLT imaging in the near-infrared (NIR) spectrum, specifically between 700–900 nm, which is particularly effective for deep tissue imaging.

By integrating this FLT technology into an endoscopic system incorporating standard white light imaging, we aim to provide surgeons with enhanced visual information. This dual-imaging capability enables better differentiation between malignant and benign tissues, which traditional white light endoscopy imaging systems cannot adequately achieve. Current systems often require separate RGB and FLT imaging cameras, leading to alignment issues and bulky configurations.

To address these limitations, we have developed a comprehensive endoscopy system that combines a rigid Hopkins endoscope with a state-of-the-art time-gated camera utilizing the CAPS sensor (tauCAM). This system capitalizes on the strengths of FLT to improve intraoperative imaging.

Additionally, we have introduced a novel deep learning-based algorithm, FLTCNN, which accurately estimates fluorescence lifetimes without requiring the system’s intrinsic parameters. The algorithm significantly reduces computational demands using a few data points, making real-time FLT imaging feasible.

Moreover, we have introduced a technique known as time-sequential RGB imaging. In this method, the surgical scene is sequentially illuminated with red, green, blue, and NIR light pulses, enabling the simultaneous acquisition of both RGB and FLT images. These images can be overlaid to give surgeons comprehensive real-time visual feedback.

Our research addresses the critical need for efficient recording of IRFs and dramatically reduces the time points required for lifetime estimation. By combining time-sequential RGB imaging with FLT in a single-camera system, we overcome challenges related to image alignment and equipment bulkiness. This development promises an advancement in surgical imaging technology, offering improved precision and better outcomes for cancer patients.

• First PhD thesis in the field of biomedical engineering (equipment and physical modelling).
• First home made thin film active electrode for electrocardiology.
• Semiconductor technology team counts 3 members.

Five young academics have been chosen to take on administrative tasks for a year in addition to their academic work to support the rector and vice-rectors. From ETRO, civil engineer Jeroen Van Schependom will take care of the vice-rectorate Research Policy.
These five academic staff members will have the opportunity, with the current management team, to develop their leadership potential and inspire the rectoral policy team. They will devote one day a week within their current tenure to this new role. Each will work closely with the rector or a vice-rector in a specific policy area to gain a tangible view of what leadership and policymaking means in practice.
“By giving young academics the opportunity to hone their policy competencies and weigh in on VUB policy, the university aims to increase its policy capability. The voice and views of our younger colleagues are absolutely essential. After all, they are also the leaders of the future,” says rector Caroline Pauwels.