Fuzzy deep LSTM for cancer cell death analysis
Researcher, under Dr. Devinder Kaur
Cell-death dynamics recovered from unlabelled microscopy video, without staining.
The problem
Studying how cancer cells die normally requires fluorescent labelling or staining. That is invasive, slow, and it can alter the biology being observed — the measurement changes the thing measured.
The question: can death patterns be recognised from raw, unlabelled microscopy video alone?
The data
Label-free, unstained microscopy video of live cancer cells — sequential frame data capturing morphological change over time.
Approach
A hybrid fuzzy deep LSTM. A fuzzy-logic layer manages noise and ambiguity in unlabelled cell imagery, where the boundary between a living and a dying cell is genuinely gradual rather than a step.
A deep LSTM models the temporal progression of cell death across frames. The architecture is novel to this application.
Results
The work demonstrated that meaningful cell-death dynamics can be extracted from label-free video, reducing dependence on staining.
Published in Neural Computing and Applications, Springer Nature.
Quantitative results are in the paper. Its figures are the publisher’s and are not reproduced here.
Limitations and what I'd do differently
Without published metrics reproduced here, the claim to evaluate is qualitative: the dynamics are recoverable. How accurately, against what ground truth, and on how many sequences are the questions a reader should ask, and the paper is the place they are answered.
Label-free imaging trades invasiveness for signal. A method that works on one cell line and one imaging setup has not been shown to transfer to another.
Artifacts
Published in Neural Computing and Applications, Springer Nature; citation to follow.