Studentka práci řešila ve spolupráci s externím pracovištěm Bruker. Hlavní milníky práce představovala vedoucímu na konzultacích. Z pohledu odborné i formální stránky nemám k práci větších výhrad a práci můžu doporučit k obhajobě. Níže přikládám hodnocení konzultanta práce.

Hodnocení konzultanta práce Ing. Michala Čičatky, Ph.D. ze společnosti Bruker, která práci zadávala:
Diplomová práce Bc. Terezy Kalivodové vznikla v přímé spolupráci se společností Bruker s.r.o. a zabývá se automatizací vizuální kontroly kvality MALDI destiček v robotickém workflow přístroje MBT PrepMatic™. Téma úspěšně kombinuje biomedicínské inženýrství s pokročilými metodami počítačového vidění a hlubokého učení. V teoretické části studentka prokazuje vysokou erudici při popisu fyzikálně-chemických principů MALDI-TOF mass spektrometrie a rešerše hlubokých neuronových sítí.
V praktické části studentka navrhla komplexní framework pro lokalizaci spotů, segmentaci naneseného materiálu s využitím pokročilých modelů řady U-Net a verifikaci depozice kapalin pomocí sítě ResNet-18. Navržené algoritmy vykazují vysokou robustnost a přesnost. Klíčovým přínosem a potvrzením kvality celé práce je skutečnost, že dosažené výsledky jsou v současné chvíli zapracovávány z prototypu přímo do řídícího softwaru přístroje a brzy tak budou nasazeny v reálné průmyslové a klinické praxi.
Studentka pracovala svědomitě po celý rok, projevovala mimořádný zájem o řešenou problematiku a velmi oceňuji, že sama aktivně přicházela s inovativními a iniciativními řešeními technických výzev. Práce je sepsána v anglickém jazyce na špičkové úrovni. Výstupy mají jasný aplikační dopad, a proto práci jednoznačně doporučuji k obhajobě. Points proposed by supervisor: 95

The submitted master's thesis addresses the automation of visual quality control of MALDI target plates within the MBT PrepMatic™ workflow. The topic is highly relevant from both scientific and industrial perspectives, as the quality of sample preparation has a direct impact on the reliability of MALDI-TOF MS analysis. The thesis focuses on the development of a comprehensive image-analysis framework combining classical computer vision methods, deep learning techniques, and quantitative quality assessment metrics. The objectives defined in the assignment were fulfilled in their entirety.
The thesis is well structured and logically organized. The theoretical part provides a sufficiently detailed introduction to MALDI-TOF mass spectrometry, MALDI target preparation, image processing, and modern deep learning approaches. The author demonstrates a solid understanding of the addressed domain and appropriately explains the principles of convolutional neural networks, segmentation architectures, transfer learning, and image classification methods. The practical part naturally follows the theoretical background and clearly describes the individual stages of the proposed solution.
From a formal perspective, the thesis is prepared at a very high level. The text is written in clear and understandable technical English, with only minor stylistic imperfections that do not affect readability. The document contains a suitable number of figures, tables, and diagrams that effectively support the presented methodology and results. The overall extent of the thesis is adequate for a master's thesis and reflects the complexity of the solved problem.
The author demonstrates very good work with scientific literature. The cited sources are relevant, up-to-date, and cover both the biomedical background and modern machine learning approaches. References are properly integrated into the text and support the presented theoretical concepts and methodological decisions.
The practical contribution of the thesis is significant. The author designed and implemented a complete pipeline for spot localization and extraction from MALDI target plate images, developed a semi-automatic annotation workflow, trained and evaluated several deep learning segmentation models, and proposed quantitative metrics describing spot quality. Particularly valuable is the combination of segmentation-based quality assessment with an additional front-camera classification module that provides information about the execution of deposition steps during sample preparation. The results achieved demonstrate a high level of technical competence. The reported segmentation performance, reaching a Dice coefficient of approximately 0.95, together with the strong agreement between automatically computed and reference quality metrics, confirms the suitability of the proposed approach for practical use.
An additional strength of the thesis is the critical evaluation of model behavior, including the analysis of false positive detections on empty spots and the discussion of potential limitations of the proposed methods. The conclusions are supported by experimental results and are formulated in a realistic manner. The presented framework has clear potential for further development and integration into industrial MALDI-TOF workflows, where it may contribute to increased reproducibility, reduced manual inspection effort, and improved quality control.
Overall, the thesis demonstrates a high level of independence, technical proficiency, and engineering thinking. The work significantly exceeds the standard expectations for a master's thesis in both scope and practical applicability.
Therefore, I recommend the thesis for defense and evaluate it with 93 points, corresponding to grade A (Excellent). Topics for thesis defence:

1. 1) The segmentation model was trained only on spots containing deposited material. How would the training strategy and dataset composition need to be modified if the system were expected to reliably distinguish between occupied and completely empty spots?
2. 2) The proposed quality metrics are based on geometric properties of the segmented regions. How do you envision validating these metrics against the actual quality of the acquired MALDI-TOF mass spectra, and which metric do you expect to have the strongest correlation with spectral quality?

Points proposed by reviewer: 93