LSU Paths to PhD: Murtaza Aslam Uses AI and Light to Advance Pancreatic Cancer Detection

For LSU PhD candidate Murtaza Aslam, the future of cancer detection may come from an unexpected combination of light, artificial intelligence, and engineering. 

His research explores how fluorescence and Raman spectroscopy can reveal hidden biochemical “fingerprints” of cancer, potentially helping doctors diagnose disease earlier and improve treatment precision.

Along the way, Aslam has combined expertise in engineering, medicine, and explainable AI to tackle one of healthcare’s most difficult challenges.

What interested you initially about your thesis research topic?

What initially drew me to my research was the opportunity to meaningfully bridge engineering and medicine. I was particularly interested in how advanced computational techniques, like machine learning, could uncover hidden patterns in biomedical data that are not visible to the human eye.

The idea that we could improve cancer detection and make clinical decisions more precisely using data-driven methods really motivated me to pursue this topic.

What are fluorescence and Raman spectrums? How are they unique for different tissues, for example, cancer vs. non-cancer?

Fluorescence and Raman spectroscopy are optical techniques that help us understand the biochemical composition of tissues without physically altering them. Fluorescence measures how tissues emit light after being excited, while Raman spectroscopy captures how light interacts with molecular vibrations.

What makes them powerful is that different tissues, like healthy and cancerous ones, have distinct biochemical signatures. Cancer tissues often exhibit altered metabolism, protein structure, and cellular organization, leading to unique spectral patterns. These differences allow us to distinguish between normal and diseased tissue non-invasively.

What is your favorite fun fact about your research, or what do you wish more people knew?

An interesting aspect of my research is working with preclinical cancer models, in which we inject tumor cell lines into mice and observe the gradual development of tumors on their backs. It is incredible to see how cancer cells grow and form tumors in a living system, which helps us understand disease progression and test new diagnostic approaches.

I wish more people knew that cancer research is not just about microscopes and data; it is also about studying how diseases behave in real biological environments. Watching tumor growth and using advanced optical techniques such as fluorescence and Raman spectroscopy to detect cancer noninvasively made me realize how future diagnostics could become faster, safer, and less invasive.

How did you use machine learning to improve cancer detection, and how can this improve chemotherapy precision?

In my research on pancreatic ductal adenocarcinoma (PDAC), we applied machine learning to both fluorescence spectroscopy using indocyanine green (ICG), a clinically approved fluorescent dye, and Raman spectroscopy.

Instead of relying on simple intensity thresholds, we extracted meaningful features from the spectra and ranked them to identify the optimal feature set, then applied machine learning models to distinguish PDAC tissue from surrounding healthy tissue. These selected features acted as key descriptors for accurate cancer classification.

This approach can improve chemotherapy precision and surgical planning by helping clinicians better identify tumor boundaries and tissue characteristics. In the future, it may support more personalized treatment strategies while minimizing damage to healthy tissue.

What were some of the biggest challenges in your project, and how did you overcome them?

One of the main challenges in my project was to define the most significant “fingerprint region” for pancreatic ductal adenocarcinoma (PDAC). Raman spectroscopy captures a broad range of biochemical information from proteins, lipids, nucleic acids, and other cellular components.

However, not all spectral regions are equally useful for cancer detection. The key was to identify which wavenumber regions offered the most reliable diagnostic signals for differentiating PDAC tissue from background tissues.

To address this, I employed explainable artificial intelligence. Rather than treating the model as a black box, I examined which Raman shifts most strongly influenced its decisions. This helped me identify spectral regions that could be crucial for PDAC classification in clinical applications.

I also aimed to connect the machine-learning results to potential biochemical signatures of cancer, enhancing the model’s interpretability and clinical relevance.

What were some of the most surprising or impactful findings? What are the implications?

One of the most surprising and impactful discoveries was that we could use the tumor-to-background ratio (TBR), a metric typically used in medical imaging, for fluorescence spectroscopy to detect PDAC.

To our knowledge, this was the first instance of applying a TBR-based method to spectral data rather than images. Nonetheless, we observed that depending solely on TBR was inadequate for dependable cancer detection.

By integrating spectroscopy with machine learning, we revealed hidden patterns that greatly enhanced classification accuracy over simple threshold methods. This suggests that future cancer diagnostics could benefit from a shift from traditional single-metric approaches to more sophisticated, data-driven systems.

What are your plans after graduation? What will you take away from your PhD experience at LSU?

After graduation, I plan to continue working in research and development at the intersection of artificial intelligence and healthcare, with a focus on translational applications that can directly impact patients.

From LSU, I will take away not only technical expertise but also the ability to think critically, work across disciplines, and translate complex ideas into real-world solutions. The experience has prepared me to tackle challenging problems in both academia and industry.