ADVANCED MICROSCOPY AND DEEP LEARNING FOR LABEL-FREE IMAGING OF CELLS AND TISSUES

LABEL-FREE IMAGING REPRESENTS A PROMISING FRONTIER IN BIOMEDICAL MICROSCOPY, ENABLING THE OBSERVATION OF CELLS AND TISSUES WITHOUT FLUORESCENT MARKERS OR INVASIVE TREATMENTS. THIS THESIS EXPLORES THE INTEGRATION OF ADVANCED MICROSCOPY TECHNIQUES WITH DEEP LEARNING METHODS TO ENHANCE QUANTITATIVE AND INTERPRETIVE ANALYSIS OF MICROSCOPY IMAGES. THE WORK FOCUSES ON THE DEVELOPMENT OF DEEP LEARNING ARCHITECTURES FOR SEGMENTATION, CLASSIFICATION, AND MORPHOLOGICAL AND BIOPHYSICAL CHARACTERIZATION OF CELLULAR AND TISSUE STRUCTURES ACQUIRED WITH BRIGHTFIELD, PHASE-CONTRAST, AND OTHER LABEL-FREE IMAGING TECHNIQUES. MANY OF THE PROPOSED METHODS INCORPORATE EXPLAINABLE AI TECHNIQUES, SUCH AS SALIENCY-BASED VISUALIZATION AND FEATURE ATTRIBUTION, TO ELUCIDATE WHICH IMAGE REGIONS AND BIOPHYSICAL CUES DRIVE MODEL PREDICTIONS AND TO INCREASE TRUST AND INTERPRETABILITY FOR BIOMEDICAL EXPERTS. THE PROPOSED APPROACHES ARE EVALUATED ON DIVERSE BIOMEDICAL IMAGING SCENARIOS, DEMONSTRATING IMPROVED SENSITIVITY AND SPECIFICITY IN PHENOTYPING TASKS, ENHANCED IMAGE QUALITY AND INTERPRETABILITY, AND PRESERVATION OF SAMPLE VIABILITY AND THROUGHPUT. OVERALL, THE THESIS SHOWS HOW THE SYNERGY BETWEEN ADVANCED MICROSCOPY, DEEP LEARNING, AND EXPLAINABLE AI CAN ACCELERATE LABEL-FREE BIOMEDICAL IMAGING, SUPPORTING APPLICATIONS IN DIAGNOSTICS, DRUG SCREENING, AND PRECISION MEDICINE.